Jennifer,
If you decide to run “3dANOVA3 -type 5” with groups, tasks, and subjects as factors A, B and C, you'd better remember that your sample size would be 1 for all cells (n=1). Thus the degree of freedom for the total error would be n-1 = 0, and you would not be able to run any F tests that involve MSE, such as main effects for Factor C and interaction effect between Factor B and C, etc. And you can't do any T tests to check those mean differences.
I agree with Doug that the best strategy for your case with current version of 3dANOVA would be to remove two subjects from the larger group, and then run 3dANOVA2 to enjoy the full spectrum of variance analysis with all statistical tests available. You might try different deletions to see which would give you more sensible results.
Gang