sendmail performance question

[David A. Sinck]

I could stand a few pointers here, as I've reached the bottom of my
idea barrel and I've got splinters from scraping the bottom.

Synopsis:  

sendmail on a dual p3/733 w/ 1G + scsi is 30% of the speed of a p3/667
w/ 512M + ide.

That just doesn't seem right; I want to know why.

If you're not interested, this would be a good time for the D key :-)


Details:

Common:
* same perl version
* same perl Net::SMTP version
* same sendmail + sendmail configs (md5sum the same)
* same input (1k test messages)
* same network (xxx.3 + xxx.4)
* same nics
* both 'mostly idle'

Slow:
* dual p3/733, 1G, scsi
* 5+m return
* kernel 2.4.9-13smp  (redhat)
 
Fast:
* p3/667, .5G, ide
* 1+m return
* local dns server
* kernel 2.4.9-12 (redhat)


The slow box uses the dns server on the fast box...and peculiarly, dig
against it comes back faster reliably based on hand tests when the dns
server is accessed remotely.  (1-2ms vs 3ms).

hdparm: buffer cache faster on slow box (scsi) (2x) but slower on
buffered disk reads (1/2) vs the fast box.

I reluectantly downed the local firewall on the slow box (albeit still
behind a router/firewall, I'm not completely insane :-), but that
didn't change the performance.

It *appears* as if the box with the local bind on it is forking
sendmail 3-4x what the non-binded box is, which would account easily
for the performance....  But if the configs are the same, how is that
possible? 

Ideas?  Preferably good ones?

TIA,

David