Mail server intermittent freeze

[Rich Winkel]

I'm pulling out what's left of my hair trying to figure out this one.
It's not a pretty sight.  Save the people who have to look at me!

It's a 1ghz intel P3 with 512MB ram, running 4.11-p26 with sendmail,
imapd-uw, qpopper, stunnel-4.14_2 and p5-Mail-SpamAssassin-3.1.1, as
well as tty logins.
It's an nis and nfs client (nfs=home dirs).  It has an fxp network card
which is in polling mode (although the problem started before it was put
in polling mode)

The system will freeze maybe every 5 minutes, sometimes for up to
a minute.  Almost completely: low level terminal io on the console
still works.  I can switch tty's (ALT-Fn) and carriage returns
are echoed (and discarded) while showing "netstat -w 1" output.
But interactive prompts are frozen.  No user-level processing is
apparent.  Network traffic is heavy, we're getting massive amounts
of spam, and spamd's load on the system has jumped considerably
in the past few weeks.  Average system load used to hover around .8,
now it averages over 2, mostly due to spamd.

Large jumps in outgoing packet activity appear with "netstat -w 1"
in the first update after recovery from the freeze.  Usually an elevated
rate just prior to it.

My main concern is the discontinuity of the behavior.   The
response is not linear with respect to load, it just drops off a
cliff, prior to which it seems normal.  I'm thinking there's a
logjam somewhere which could be alleviated.  

/etc/sysctl.conf contains

net.inet.icmp.icmplim=100
kern.ipc.somaxconn=1024
net.inet.tcp.icmp_may_rst=0
net.inet.ip.redirect=0
net.inet.tcp.blackhole=1
net.inet.udp.blackhole=1
kern.polling.enable=1

Has anyone else seen this behavior??

Thanks for any help!!

Rich