While interactive response (short single packets) was really "snappy"
(very fast). The longer transfers seemed to "pump" a lot (the rate
going up and down). I don't know if that behavior was caused by the
remote end's process scheduler or something similar, but it might be
some indication that TCP flow control is not optimal. I think a more
gradual change in the parameters would be good (if Linux is
responsible for that). Sorry, this is a shot in the dark!
Ulrich