System perforamance 4.x vs. 5.x and 6.x

[Brett Bump]

I've recently upgraded a mailserver from a 4.x version to 6.2.
This server had been upgraded a few years ago to 5.x but the
performance was so bad that we only let it run a few days before
moving it back to 4.x.  Years pass and it seemed time once again
to move forward.

What is the magic bullet in getting the same kind of performance
out of a 5.x or 6.x version that I've just come to expect from
FreeBSD ever since version 1?

I'm seeing signal 6's on apache and imapd (never happened before)
network errors, serious response time errors and generally poor
performance during peak activity (same box, same people).

ufs memory looks exactly like it did before and doesn't max:

vfs.ufs.dirhash_minsize: 2560
vfs.ufs.dirhash_maxmem: 2097152
vfs.ufs.dirhash_mem: 1923157
vfs.ufs.dirhash_docheck: 0

mbufs hasn't changed:

536/604/1140 mbufs in use (current/cache/total)

and disk performance is very good EXCEPT during peak activity:

--------------------------------------------------------------------
Mail Server (Dual Xeon P4 3mhz 2g memory [Perc] U320):

-bash-2.05b$ time dd if=/dev/zero bs=1024k of=tstfile2 count=1024
1024+0 records in
1024+0 records out
1073741824 bytes transferred in 47.037099 secs (22827552 bytes/sec)

real    0m47.041s
user    0m0.000s
sys     0m5.444s
-bash-2.05b$ time dd if=tstfile2 bs=1024k of=/dev/null
1024+0 records in
1024+0 records out
1073741824 bytes transferred in 2213.643946 secs (485056 bytes/sec)

real    36m53.647s  <---Check it out.
user    0m0.008s
sys     0m3.619s

--------------------------------------------------------------------

I've changed the order of php extensions, disabled autonegotiation,
moved mail queues and large volume directory folders to separate
drives and set noatime.  Nothing seems to make much of an impact.
My next idea was to setup my kernel for device_polling, but none of
this is really diagnosing what the real problem is.  Any clues?

Brett