Please tar up and e-mail me all the input files (except the AFNI datasets) needed to reproduce this problem -- including the offending 3dDeconvolve command. And tell me how many points in the time series data (e.g., output of 3dnvals on the input dataset).
I don't want the whole afni_proc.py stuff, since that seems to be what you are using. Just the part that is giving trouble!