James,

I don't know what's going on, but you really have some weird input files. Basically all those subjects don't vary much in terms of the contrast between the two conditions compared to the within-subject variability, and you can see that both tau^2 (sub-birck #2, cross-subject variance) and Qe (sub-brick #3, chi-square test for Ho: tau^2 = 0) in both of your output files are all 0's across the brain, which means that the random-effects model is essentially equivalent to a fixed-effects model! And in such a special scenario t-statistic is definitely more appropriate than Z.

You can run 3dANOVA2 -type 3 on your data, and the result should be similar to 3dMEMA with t-statistic as output.

This makes me think that maybe I should just stick with t-statistic and dump Z? Hmm...

Gang