I guess the antialias filter starts rolling at around
20, and uses all
the extra bandwith to avoid making the filter steeper and messing the
'audible' audio. Just a guess. I would guess most music recording
interfaces do this.
I realized why they do this. With variable sampling frequencies,
44.1, 48, 96, you would need different filters to get the best out of
each one. So they just do 20 once and throw away the extra bandwidth.
So does that mean they will alias at sampling rates like 11025? This
could be tested, too.