For each subject the residuals from the "list.of.normal.control.ERRTs.files+tlrc" is a time series, while the corresponding beta from that subject in "list.of.normal.control.fear.files+tlrc" would be just one number. If you try the following
3dttest -set1 `cat list.of.normal.control.ERRTs.files+tlrc` -set2 `cat list.of.normal.control.fear.files+tlrc` -prefix Controls.Fear_Versus_Errts
you will get the same result as your option 1 because each subject's residuals add up to 0 in this two-sample t-test.
> Our concern with using "-base1 0" is that wouldn't this t-test only identify
> voxels whose Coefficient are significant from 0, not neccessarliy different from
> a baseline (or fixed)?
Since the baseline is automatically added in the regression model (unless you explicitly turned it off with something like -polort -1 in 3dDeconvovle), each regression coefficient (beta) is indeed a contrast relative to the baseline by default.
Gang