CUPS error_log says Max Clients Reached,cups slow

[ed smith]

About once every 3 weeks, users complain that "CUPS is slow".(Other factors eliminated),  the 2 indicators that always occur when this happens are:

1) CPU time for the CUPS daemon runs higher than normal.
2) The /var/log/cups/error_log has the following warnings/errors:

E [16/Nov/2005:04:08:29 +0600] Unable to accept client connection - No buffer space available.
E [16/Nov/2005:04:12:43 +0600] Unable to accept client connection - No buffer space available.
E [16/Nov/2005:08:33:29 +0600] Unable to accept client connection - No buffer space available.

AND

W [16/Nov/2005:10:29:10 +0600] Max clients reached, holding new connections...
W [16/Nov/2005:10:29:10 +0600] Resuming new connection processing...
W [16/Nov/2005:10:29:10 +0600] Max clients reached, holding new connections...
W [16/Nov/2005:10:29:23 +0600] Resuming new connection processing...
W [16/Nov/2005:10:29:23 +0600] Max clients reached, holding new connections...
W [16/Nov/2005:10:29:23 +0600] Resuming new connection processing...
W [16/Nov/2005:10:29:23 +0600] Max clients reached, holding new connections...
W [16/Nov/2005:10:29:26 +0600] Resuming new connection processing...
W [16/Nov/2005:10:29:26 +0600] Max clients reached, holding new connections...

appear every 30 seconds. normally they don't.

I had success once with bumping maxclients from 500 to 800 in /etc/cups/cupsd.conf, and then doing a lpshut, lpsched. (We have 200 printers on a single HP/UX 11.0 server). Since that time, I havent been able to alleviate the situation when it occurs by increasing or otherwise manipulating this nerd-knob.

Recycling CUPS always works, but this is intrusive and I would like to resolve the problem. I saw the other posting listed on this subject.
I checked ulimit, and it says its UNLIMITED. Is there another kernel or CUPS parm I should investigate?

any help appreciated.
Thanks!