Actually, I do not know the SPM VBM pipeline well. Since in FSL the default is permutation test, for the users of pTFCE, I decided not to recommend the current, GRF-based pTFCE (and any GRF-theory based approach) for VBM. But of course, if you have a good reason to believe that GRF-assumptions approximately apply to your data then, I think, you are free to use pTFCE, as well. Actually, it might be the case for the SPM VBM pipeline...
Do you maybe know some reference regrading this (the use of GRF-based thersholding in SPM for VBM)? I might have a closer look later...
Or maybe someone else on the list could help us with this...
I have dowloaded your files, it is indeed interesting how FSL TFCE seems kind of “overboost” signal.
Could you tell me the T-score threshold for the spmT image (FWER p<0.05)?
Further, we could have a look at the non-TFCE output of randomise (-x option of randomise). Do differences originate from the TFCE-approach of the permutation-testing?
(BTW: please double check you randomise run, e.g. number of permutations)