For spamassassin to work well, make sure bayesian filtering and auto-learning are on. You should also lower your spam threshold to around 3.5 or so. After 2 weeks or so, enough mail will have been processed so the bayes database works well (I think it takes ~300 emails, you can speed this up using sa-learn on any existing email), and auto white listing will help eliminate any false positives.
You should also set up a folder where you can drop missed spam to have spamassassin learn from it, that has helped me tremendously and is easy to use since you just drop the messages in the folder.
I can post my config files and the cron entry to learn from missed spam if you want.