Logged In: YES 
user_id=46639

I've been unable to repeat the problem through a tcpwatch.py 
proxy, so I'm guessing the trigger is connecting to a fairly loaded 
server over a 56k modem - possibly the socket is in a bad state 
and nothing noticed?

I'll try not going through tcpwatch.py for a bit and see if I can still 
trigger the problem in case there was a server problem triggering 
it that has been fixed.