OK, I think I've found the issue: there was a variable in 3dRSFC initialized to be "frequency greater than Nyquist" in order to provide an upper limit of the "full band" denominator in the fALFF calculation. This number was very big, so dividing by such a small df (=1/(TR * Nt), where Nt was very large in this time series), produced a numerical error. I've set the initialized frequency to be smaller, but still greater than Nyquist, so this problem won't occur.
I've push in the changes, so in the next update, this problem should be solved. Let me know if there are further issues with it.
--pt
ps: that's a really long duration time series-- what patient subjects you must have!