At the moment, my thought would be that processing with afni_proc.py is a good thing in terms of the overall data; the ability to include the RSFC parameter estimations directly in this would be useful, yes.
I don't think one can use 3dRSFC on data which has already been LFF-ified, since parameters like fALFF will need un-bandpassed spectra for comparison.
In afni_proc, I believe that the bandpassing is done as a regression, just at the very end, while lots of other things are being regressed. I don't know if it would be a problem to *not* do the bandpassing here, but leave that for after afni_proc.py using 3dRSFC on the former's output errts file? It would mainly be a question of whether it's a problem or not to do the filtering after everything else. When using 3dRSFC then, one would *not* want to blur/detrend/anything but bandpass.
I'm afraid I don't have a script for putting together 3dRSFC and afni_proc.py... An example way of running 3dRSFC could be:
$ 3dRSFC -nodetrend -mask MASKNAME -bp_at_end \
-prefix PREFIX \
0.01 0.1 INPUT_NAME
(this would only do bandpassing, such as if you were running after a mostly-full afni_proc.py and would give values of ALFF/fALFF/RSFA and the filtered time series.)
In terms of papers, I think it really depends on the questions you are asking in the study, and also what kind of population is being used. I would assume there would be pretty full coverage of permutations of RSFC parameter and subject population... NB, you can also use 3dReHo to calculate ReHo on a final output LFF set.
--pt