Hello,
1) It looks like updates have been made regarding that permute failure, so I would be inclined to update your AFNI package and run it again. The most recent updates were made this past March (2022). There is a reasonable chance that something has been fixed in that regard.
2) Do you mean that 3dttest++ -Clustsim required clusters to be around 142 voxels while 3dClustsim required only 42? A size of 42 seems suspiciously low, though it depends on how large the voxels are.
Those are very different methods, one based on permutation tests of actual brain data, the other based more on blurred random noise.
The -Clustsim approach is easier to defend. I won't presume to attach the word "better".
- rick