As an observer, I have to wonder what the heck happened in the
design, development and testing of 5.4.

If you can't do a trace, a cheap substitute is to sample what's going
on with Query SEssion. and Select * From Sessions.  The Select is
helpful in showing data progress down to the byte level, plus the in
progress verb, for a better sense of what's going on.  The session
state can help identify where the delay is.  You might try an
artificial test, with a single, very large file, then a large set of
small files, to see if basic transmission speed or transaction rate
is the problem.  A foul-up in TCP Window Size could be the
impediment.  If you have the time, try a few different key settings
and see if an affector stands out.

   Richard Sims