I don't know if it matters but I'll throw this out anyway, maybe it will help someone. I don't run bayes, I don't run autowhitelist, I don't run pyzor/dcc/razor. I have never seen these issues of high memory/cpu useage. I amd handling 700/1300 messages per hour ( we are an ISP, traffic levels change daily 8^), 60% of which is spam this afternoon.
This is done with three machines using spamc to connect to a master server running spamd with user prefs stored in MySQL. spamd is running with the following startup under daemontools,
#!/sbin/sh
PATH=/usr/bin:/usr/local/bin
exec /usr/local/bin/softlimit -a 128000000 \ /usr/local/bin/spamd -i 10.0.240.253 -p 1783 -A 10.0.240.0/24 \ -m 10 --max-conn-per-child=200 -u vpopmail -x -q -s stderr 2>&1
DAve
-- Systems Administrator http://www.tls.net Get rid of Unwanted Emails...get TLS Spam Blocker!