Sounds like you did a two-tailed test. shminux's hypothesis, which he has stated several times IIRC, is that people who can solve it will not be taken in by Eliezer's MWI flim-flam, as it were, and would be less likely to accept MWI. So you should've been running a one-tailed t-test to reject the hypothesis that the can-solvers are less MWI'd. The p-value would then be something like 0.13 by symmetry.

Yes, but I imagined someone like Eliezer might have the hypothesis that the math naturally leads to MWI and rationalists who understood the math would realize that.