[Mimedefang] sa-learn load, blocking?

[Adam Porter]

Apologies if slightly off-topic, I'm guessing this is mainly a 
spamassassin thing though I'm not positive.

I'm using Redhat Enterprise Linux 3 on a 2.4Ghz Xeon with 2GB of RAM, 
Mimedefang 2.41, McAfee (uvscan) and spamassassin 2.63.  We use global 
bayes learning like this:

use_bayes 1
bayes_auto_learn 1
bayes_file_mode 0770
bayes_path /var/spool/MIMEDefang/bayes
bayes_auto_learn_threshold_spam 8.0
bayes_learn_to_journal 1
bayes_auto_expire 0

Once a night, this gets run out of cron to synchronize the database:

/usr/bin/sa-learn -p /etc/mail/spamassassin/sa-mimedefang.cf --force-expire

The problem comes when trying to feed the database ham and spam this way:

/usr/bin/sa-learn --showdots --spam --dbpath /var/spool/MIMEDefang/ \
     /home/$person/Maildir/.spam/cur  --no-rebuild

(or --ham with ham folder instead of --spam with spam folder.)

We find that the sa-learn seems to lock something (the database?) and 
when feeding sa-learn hundreds of messages at a time, the load average 
skyrockets and the machine starts to reject inbound messages due to 
sendmail's threshold (already raised to 30)!  (RAM usage does not seem 
to be a problem in this case.)

I'm thinking of upgrading (both spamassassin and Mimedefang) but would 
feel better about the downtime involved if I knew this problem would go 
away.

Suggestions welcome!

Thanks,
-Adam