Please, see at this link a mockup design for 7 subjects, being 4 in one group, 3 in the other. The 8 conditions are named from A through H: https://dl.dropboxusercontent.com/u/2785709/outbox/mailinglist/design_graham2.ods

The contrasts that are stricken through aren't meant to be run, and are here only to show how the others were constructed.

With a 2x2x2 experiment we have then 8 conditions that can be all compared to each other (within-subject), plus across groups (the fourth factor, between-subject), plus the interactions, all positive and negative directions. Overall, 226 tests are possible using the same data, plus 5 F-tests. Many of these certainly don't make sense in the real world, but regardless, there will still be many tests possible. Thus, correction across contrasts is essential.

I suggest using permutation tests, with exchangeability blocks, as available in PALM. There will be one EB per subject, with the permutations happening within-block(for the within-subject effects) and simultaneously whole-block (for the between-subject effects). Please see the documentation and if something isn't clear, feel free to ask again.

All the best,

Anderson