I've always felt it was an urban myth too, but every time I've said so on a forum, someone inevitably comes back with "it worked for me", so I've given up arguing the point.
My suspicion is that when it does *appear* to work, the drive may have temporarily staggered back into life for some other reason, and that any success was in spite of the freezing rather than because of it. It's hard to make an argument with figures though, as people are much less likely to report "My drive died, I put it in the freezer, and it was still dead afterwards", even though they may far outnumber the alleged successes.
I guess you could run an experiment, but you'd need access to a large number of failed drives, enough to provide a meaningful sample size - then you could freeze half of them, and see if the recovery rate was any better than the non-frozen control group (which could simply be left powered down for the same length of time at room temperature). Personally I wouldn't be surprised if the recovery rate was *lower* for the frozen drives, as it's difficult to imagine how it wouldn't do more harm than good under any conceivable circumstances.