Thank you very much for your quick and helpful response!
I had not realized that the current GRF-based implementation of pTFCE was not appropriate for VBM analyses, which is what these are. My understanding was that the application of a smoothing kernel to the images (we use 8mm FWHM) helped to satisfy the assumptions required for GRF-based correction, and that as such SPM's native GRF-based FWE correction is commonly used for VBM and considered appropriate (albeit overly conservative). Please correct me if I'm missing the distinction here!
Would you be able to say then whether it makes sense that we would see many more / larger clusters across the brain with FSL's permutation-based TFCE than with pTFCE? (perhaps this also touches on your 3rd point?) The differences were just so striking that it seemed hard to believe that the former were all true positives. FWIW here is a link to our *_tfce_corrp_tstat* and *spmT* images (Degrees of freedom is 776): https://www.dropbox.com/sh/qwo4evr45k06f0e/AABXDAHRUeBI6Krwzm344pmZa?dl=0
Thanks again for your help and for your work on this tool!