COFFEY, Circuit Judge.

On April 30, 1990, Clifford Olson filed a habeas corpus petition pursuant to 28 U.S.C. § 2255 and a motion requesting an evidentiary hearing. Each of Olson's motions sought to challenge a murder conviction on the basis of "newly discovered evidence." The United States district court converted his habeas petition into a Federal Rule of Criminal Procedure 33 motion for a new trial based on newly discovered evidence because he filed it within two years of the final judgment against him for the 1977 murder of Clifford George Albers on Indian territory. On November 16, 1991, the district court denied Olson's Rule 33 motion for a new trial and his request for an evidentiary hearing. Olson has filed a timely appeal to the district court's judgment. We affirm the district court.

I. FACTS

A recitation of the factual background in this case is essential to understanding the present litigation. In 1980, a federal grand jury indicted Clifford Olson for the gruesome 1977 murder of Clifford George Albers on the Menominee Indian Reservation in Eastern Central Wisconsin. The court, on the government's motion, dismissed the indictment without prejudice shortly after filing. After a second grand jury reindicted Olson in 1985, he was convicted on September 13, 1985, before a jury for the murder of Albers.

Wanda Dick, Ella Peters, and Brenda LaRock testified for the government against Olson. Their testimony is set forth in detail in our opinion affirming Olson's conviction in United States v. Olson, 846 F.2d 1103 (7th Cir.), cert. denied, 488 U.S. 850, 109 S.Ct. 131, 102 L.Ed.2d 104 (1988). The common thread in their testimony reveals that Olson, Peters, LaRock, and Robert Kakwitch were driving down a little used road and encountered Albers who was driving in the opposite direction. Olson forcibly removed Albers from his car for no apparent reason and forced Albers to accompany him and the others into the woods where they bound and gagged Albers and forced him to sit nearby while Olson, Peters, LaRock and Kakwitch drank alcoholic beverages around a fire. Thereafter, Olson hit Albers' ankle or leg with an ax and then shot him in the face. Peters, LaRock, and Kakwitch then alternately shot Albers. At this point their respective testimony diverges:

Dick recalled at trial that Olson shot Albers in the face, and then handed her the gun and she shot Albers in the leg; she could not remember who else shot Albers. Olson directed Dick to get a blanket from the car and then they rolled Albers' body in the blanket and placed it in the trunk of the car. They then drove to the Keshena Bridge and dumped the body in the river.

LaRock testified that she did not see the actual shooting because she was a short distance away in the woods. She said that when she came out of the woods she observed Peters, Olson, and Kakwitch with guns. Dick shot Albers and then Olson gave LaRock the gun and she in turn shot Albers. She also testified that another person, Charlie Peters, was at the scene.

Ella Peters, while not referring to the presence of Charlie Peters, testified that everyone at the campfire shot Albers but that there was only one gun.

II. DISCUSSION

Olson now claims that he is entitled to a new trial (or in the alternative, a hearing to determine if he is entitled to a new trial) based on the sworn affidavits of Brenda LaRock and Ella Peters in which they recant their prior testimony inculpating Olson. This is the second time LaRock has recanted her testimony. She previously recanted in an affidavit dated August 18, 1986, roughly eleven months after Olson's trial. The district court denied Olson's motion for a new trial based upon the initial LaRock recantation. We affirmed the conviction and likewise denied the motion for a new trial through application of the test set forth in Larrison v. United States, 24 F.2d 82 (7th Cir. 1928). See Olson, 846 F.2d at 1112. In Olson's present Rule 33 motion for a new trial, he offers not only LaRock's second recantation but also the recantation of Peters. Olson argues that considered together, the testimony of LaRock and Peters constituted a substantial portion of the government's case against him, therefore, he is entitled to a new trial. The district court considered the LaRock and Peters recantations separately because it could not reevaluate the LaRock recantation since this court had already specifically denied a retrial based on the first LaRock recantation. Specifically the court stated:

"When an appellate court considers and rejects a claim on direct appeal, that decision is binding on the district court in ruling upon a subsequent motion for a new trial under Rule 33 or upon a motion to vacate a sentence pursuant to 28 U.S.C. § 2255. See Page v. United States, 884 F.2d 300, 302 (7th Cir. 1989). Therefore, this court must be bound by the holding of the court of appeals regarding the recantation of Brenda LaRock Webster."

Mem. op. at 7. The trial court faced Peters' recantation for the first time and also rejected it as not meriting a new trial. Olson appeals the denial of that motion.

The trial court has discretion to grant or deny a new trial under Rule 33. United States v. Kamel, 965 F.2d 484, 491 (7th Cir. 1992). In Larrison, 24 F.2d at 87-88, this court established the test for granting a new trial based on recanted testimony. The Larrison test was restated in United States v. Leibowitz, 919 F.2d 482, 484 (7th Cir. 1990), cert. denied, ___ U.S. ___, 111 S.Ct. 1428, 113 L.Ed.2d 480 (1991), and requires that a defendant is required to demonstrate three things before the granting of a new trial: "that the recantation is true; that the jury might have reached a different result if the witness in question had testified truthfully; and that the witness's false testimony took the defendant by surprise." Id. This court has not seen fit to apply this test rigidly in all cases, but, as we will see, the Larrison test governs the recantation proffered in this case. See United States v. Mazzanti, 925 F.2d 1026, 1029 n. 2 (7th Cir. 1991) (tracing our recent application of Larrison).

The first inquiry under Larrison focuses on whether Olson has demonstrated that their recantations are trustworthy. The history of this case recounts the history of LaRock's and Peters' flip-flops. They both testified before a grand jury that they had not witnessed the murder, shortly thereafter they changed their testimony and told a second grand jury as well as the petit jury that they had in fact witnessed the murder. During Olson's trial the jury heard these varying stories and the witnesses' explanations for the variations; it chose to accept the witnesses' version of the testimony that they had in fact observed Olson shoot Albers. It is a truism that our courts treat recantations with skepticism. E.g., Kamel, 965 F.2d at 494 n. 25; Leibowitz, 919 F.2d at 483. This case is no different. By recanting, the witnesses are simply trying to change their trial stories back to their original grand jury stories — the same stories that the petit jury heard in cross-examination and rejected. To give Olson a new trial on something that one jury already has decided would not put any important new facts before the jury, it simply would give Olson another roll of the dice. Because this court has previously ruled on the question of Brenda LaRock's recantation, it is inappropriate to revisit that question. See e.g., Wozniak v. County of DuPage, 845 F.2d 677, 680 (7th Cir. 1988) ("Res judicata is designed to ensure the finality of judicial decisions because it `encourages reliance on judicial decisions, bars vexatious litigation, and frees the courts to resolve other disputes.'" (Quoting Brown v. Felsen, 442 U.S. 127, 131, 99 S.Ct. 2205, 2209, 60 L.Ed.2d 767 (1979))). We must not permit meritless challenges to valid final judgments. Res judicata and the Larrison test balance the need for achieving justice with the necessity of preserving final judgments.

The defendant argues that the district court erred because it failed to consider the recantations together. We disagree for these recantations are dubious even when considered jointly. First, in LaRock's initial "recantation," she stated only that she could not recollect the events of the night of the murder. She did not say anything to exculpate Olson such as providing an alibi or identifying someone else as the murderer. Second, throughout Olson's first motion for a new trial, Peters never came forward to recant her trial testimony. Peters did not recant until December 14, 1989, more than four years after her trial testimony and over a year and a half after this court noted that her refusal to recant supported the jury's determination of guilt — facts very similar to those that raised the court's eyebrows in United States v. Provost, 969 F.2d 617, 620-21 (8th Cir. 1992), cert. denied, ___ U.S. ___, 113 S.Ct. 986, 122 L.Ed.2d 139 (1993). Finally, LaRock and Peters now seem to be trying to fill the void that the court relied on when upholding Olson's original murder conviction. While the present recantations highlight the inconsistencies that have existed throughout this proceeding, the inconsistencies fail to establish that the witnesses' most recent story is true, they demonstrate only that LaRock and Peters perjured themselves at one point or another. We are of the opinion the district court did not abuse its discretion in finding the trial testimony more credible and refusing to grant a new trial on this newly manufactured testimony. See Lindhorst v. United States, 658 F.2d 598, 602-03 n. 5 (8th Cir. 1981), cert. denied, 454 U.S. 1153, 102 S.Ct. 1024, 71 L.Ed.2d 309 (1992). We are persuaded, in part, by the fact that the same district court judge who denied Olson's most recent new trial motion presided over his trial and was able to view, weigh and analyze firsthand not only the witnesses' composure or lack thereof, but their testimony as well.

The trial court also found that, even without the testimony of LaRock and Peters, Olson has failed to establish that a new trial might have a different outcome. Olson says that a new trial would lead to an acquittal because now the government would only have Dick's testimony and the physical evidence used at trial. This evidence, according to Olson, would fail because it is the same evidence the government had in 1980 when it dismissed the first indictment, but interestingly nothing in the record reflects the government's reason for moving to dismiss the initial indictment. Nor does the record indicate that the government's evidence was limited to Dick's testimony and the physical evidence of the crime. Indeed, even if the government had only Dick's testimony and the physical evidence it would have in all probability been able to secure Olson's conviction. Several factors support this reasoning. Initially it is unclear that the testimony of LaRock and Peters carried much weight in convicting Olson. All indications are that Peters' and LaRock's testimony came off at best as suspect but in all probability incredible. Part of Olson's defense was to impeach their trial testimony with their conflicting grand jury testimony. Olson even introduced a lie detector test showing that Peters' grand jury testimony was true, which, whether accurate or not, must have at least strengthened the impression that Peters was incapable of giving testimony worthy of belief. Wanda Dick's testimony, on the other hand, was fairly consistent considering all aspects thereof including the period of time between the crime, the first grand jury, and the trial. Most importantly, the jury heard all versions of the story and convicted Olson, discrepancies notwithstanding. The court did not abuse its discretion in ruling that a new trial would not lead to an acquittal.

The court also found that the witnesses' testimony should not have taken Olson by surprise. Leibowitz criticized dogmatic adherence to Larrison's inquiry into whether the defendant had reason to be surprised by the testimony of the recanting witnesses. 919 F.2d at 484. Leibowitz, however, agreed that "[s]urprise is relevant. . . . If the defendant had every opportunity to meet the allegedly false testimony at trial, his failure to unmask its falsity at that time is some evidence that the testimony was true." Id. We need not rely solely on Larrison's surprise inquiry, but this is a case in which lack of surprise bears on the truth of the recantation. The lack of surprise is evident from the fact LaRock and Peters testified before the Grand Jury in February 1985 and gave the same testimony at trial in September 1985. Accordingly, Olson's trial counsel had six months to prepare cross-examination of the testimony. Thus Olson has previously had the veracity of the witnesses tested in the jury's decision-making process. As noted earlier, if the court were to accept the recantations any newly impaneled jury would face the same question as the last jury as far as LaRock and Peters are concerned: Which version of their stories do you believe? Olson has failed to present us with any new evidence that would answer this question to any greater degree of satisfaction.

Finally, Olson argues that at the very least the court should order a hearing to allow him to present his new evidence. He argues that the district court incorrectly applied the test for granting an evidentiary hearing, see United States ex rel. Simmons v. Gramley, 915 F.2d 1128, 1139 (7th Cir. 1990), but a court has broad discretion on whether to grant a hearing on a motion to introduce new evidence in the form of recantations. United States v. MMR Corp., 954 F.2d 1040, 1046 (5th Cir. 1992). Courts usually will not conduct a hearing unless there are unique circumstances. Id. A hearing is particularly rare when the recantations consist of sworn affidavits by witnesses who testified at trial before the judge who is being asked to rehear the recanting witnesses. Provost, 969 F.2d at 619-20; see Kamel, 965 F.2d at 491. Examples of unique circumstances include jury tampering, prosecutorial misconduct, or third party confessions. MMR Corp., 954 F.2d at 1046. The flip-flopping of testimony and the lengthy period of time that has elapsed since the crime have left questions open in this case, as can happen in any case. The only question not presented to the jury, however, is the one now raised in the affidavits about FBI agents pressuring witnesses into perjuring themselves. Olson, though, offered those affidavits claiming now that LaRock and Peters are reformed liars. The trial judge properly chose not to believe this in the exercise of his discretion. We agree with the trial court's ruling in denying an evidentiary hearing.

As previously mentioned, there is a need for finality in criminal adjudications. The reliability of Brenda LaRock's and Ella Peters's testimony has already been thoroughly considered by a jury, and in the case of LaRock's recantation this court has already ruled once. This court has on prior occasions stated:

In reaching our decision affirming the district court, we are mindful of the need for finality of criminal convictions. . . . All too often, as we are seeing in this case, the defense counsel takes a buckshot approach hoping a pellet will strike — but instead, our overburdened judicial system has been put through yet another needlessly repetitious and wasteful collateral attack. Some place, somewhere, somehow, we must put an end to this repetitious and meritless litigation if we are to be able to attempt to render justice in a timely fashion to those with meritorious claims.

United States v. Kovic, 830 F.2d 680, 692 (7th Cir. 1987), cert. denied, 484 U.S. 1044, 108 S.Ct. 778, 98 L.Ed.2d 864 (1988). A jury has convicted Clifford Olson of committing a gruesome murder, it is now time to cease debate over that jury's verdict.

AFFIRMED.