The average tpattern time works fine; it tends to be
approximately in the middle of the TR (slightly less
than that). You are right, the average tpattern does
not depend on the acquisition. I do not see why it
should. Bob made the default at the midpoint so
that the average time shift is minimized.
Note that afni_proc.py uses a default of the beginning
the TR, since that generally makes the timing align
with the stimulus timing.