Mark Stock's work on rendering large models w/o ambient cache

Hi,

Mark did some work on how to get nice results without ambient cache (afaik
he uses -aa .00 -d 6 -ps 1 and downsamples /4 in pfilt). I tried the same on
a very large model, but rendering is slow now because the different views
can't share ambient data. So my suggestion is to keep low -ad and downsample
as proposed, but cache the fewer ambient values in an ambient file using a
moderate -aa. Any comments? I wondered why this was not mentioned in the
studies, am I completely wrong or is it because he simply did not have to
reuse the ambient values (in a comparision with different ambient settings,
that would be a bad idea ;-)?

TIA+CU Lars.