Jeremy,
I assume that *.hit.Gb3+tlrc and *.miss.Gb3+tlrc for each subject are two subbricks from your glt output (area under the curve for hit and miss respectively) of 3dDeconvolve. If this is the case, the failure to detect any significant group difference betweem hit and miss could be due to the following three reasons:
(1) Eight subjects might be too few for group analysis with random effect.
(2) Even if you see significant differences with the output from 3dDeconvolve for each subject, they might not share much overlapping area among most of the eight subjects. If this is the case, group analysis would unfortunately fail.
(3) I imagine that there might have some nonzero covariance between hit and miss for each subject. So you might consider running 3dttest on individual coefficients of hit and miss from 3dDevonvolve. Run 3dttest with the '-paired' option. If the correlation between hit and miss is positive, the contrast test in 3dANOVA2, which is an unpaired t test, tends to be less powerful on difference detection than 3dttest with a paired contrast test.
Gang