* How Spam filtering / learning works?
* Can user add custom rules? Learn filters?
If you run GO on Debian (which i recommend), you should be able to use spamassassin and clamAV (in my case, i baked a "ClamAssassin" for my servers
I have a cron job running each hour which learns with bayesian filters, works like a charm.
You just have to educate your users NOT to delete spam, instead move it to a special folder (in my case called "training").
Unfortunately, some of my mail accounts have a little nice SPAM icon, and some don't.