SETUP A:
-a levels --> 8 one for each timepoint that you estimated in the deconvolution
-b levels --> 2 one for each task
-c levels --> 10 one for each subject
fixed, fixed, and random effects for the levels, respectively
SETUP B:
-a levels --> 8 one for each timepoint that you estimated in the deconvolution
-b levels --> 2 one for each task
fixed and fixed effects for the levels, respectively
In setup B with 3dANOVA2, you are testing whether there is any significant difference of various factor effects, among the 8 levels of factor A and between the 2 tasks. The data from 10 subjects are treated as 10 repeated observations for each combination.
In setup A with 3dANOVA3, in addition to the above difference, you view those 10 subjects as a resprentative group (random smapling) of some imaginary population. However, if you don't have any repeated scans for each combination, there is no way you can test the variability of the subject mean since the defree of freedom for MSE would be zero.
In your situation, setup A does not provide you any more statistical testing choices than setup B. Plus, you lose more statistical power with setup A since there is basically one sample for each cell. Unless you do have repeated scans for each subject (in which you can test the subject variance and interactions between subject and any of other factors), you'd be better off with setup B.
Yes, 3dANOVA4 would be very handy in some situations. I'm still working on it. Just keep your fingers crossed.
Gang