Web Server race problems with RH 7.1/Apache IMP 2.2.6

[Walter R. Moore]

Since everytthing tested well, I upgraded our webmail server from IMP 
2.2.4 to 2.2.6 and from horde 1.2.4 to horde 1.2.6. A disastrous new 
behavior has started: after a while under load Apache begins to spawn 
more and more children, all chewing up cpu cycles. The load rises 
arithmetically well before I run out of idle cpu time. I just saw it hit 
60. Stopping apache, killing the bad children and starting it makes 
things work for 30 minutes to an hour under load, then boom again.

This did not happen under the same loads with 2.2.4. Any suggestions?

Dell PowerApp.web 120 - PIII 866, 512mb memory

RedHat 7.1
php-4.0.5-2
postgresql-7.0.3-8
horde 1.2.6
imp 2.2.6
apache-1.3.19-5

I have KeepAlive turned off (I tried on as well, just to see if that 
made a difference). Web responses are quick until the load rises past 
25, there do not appear to be a lot of postgres connections hanging 
open, and disk & cpu load are low for a while.

We do about 75,000 hits/1500-2000 logins a day on the webmail server, so 
there should, in theory, not be undue stress on the box.

Anyone else solve or experience this? What would you tune? HELP! :)

-Walter Moore
  ITS

-- 
+-----------------------------------------------------------------+
Walter R. Moore --  System Administrator, Eckerd College
moorewr@eckerd.edu --  http://home.eckerd.edu/~moorewr

"If the United Kingdom is to be overun, keep outside the ring. Go to
South  Africa, Australia, Canada. Keep going, and stay in touch with
Auxiliary units in the UK. Remember, it took the Greeks only six
hundred years to get free of the Turks."
    - Peter Wilkinson, chief of staff to Col. Gubbins, SOE, May, 1940.