CPU load spikes when Spam Assassin bayes and autolearn on

[GivePeaceAChance]

At my workplace (an ISP), we're using Spam Assassin and Clam AV running on gentoo for the mailserver.

We're running an unstable version of SA, 3.2.3. The latest unstable is 3.2.4, and the latest stable is 3.2.1-r1.

For some reason, when we enable bayes and autolearn, the CPU load spikes on our server. It has two, dual-core 2 GHz Zenon processors and 4 GB of RAM.

The mail server accounts for about 2500 users.

Any idea as to why this is happening and/or how to fix it? Is it due to the instability of the version we're running? Poor coding or something? Or is it fixable? (And if it is, how can I fix it).

We *might* look at going back to the stable version, but my co-worker (who's been here way longer than I) is concerned about breaking something if we downgrade.

Thanks for your time.
_________________
I am not a Linux Guru, but a n00b with a lot of questions.