Mozilla should put a Bayesian spam filter as a high priority.
After reading the article A Plan for Spam I got fairly excited about this new method of spam filtering. Assuming that Bayesian filtering works as well as the author of the article claims, this new method works about as well as the highly regarded Spam Assasin. This method however has several important advantages. Bayesian filtering does not rely on any other machines. It makes no extra connections, no extra requests. All the processing and rules are client side. The rules in a Bayesian filter are adaptive, they adjust automatically as new emails come in.
This could be the “Outlook killer” feature that would push many to switch to Mozilla as their email client. True the user still has to download the message before being able to filter it and for many not downloading the spam message is the whole point. Possibly in the future a standard could be developed wherby a client machine periodically communicates to the server a users personal keyword ratings and filter thresholds. Then the filtering can be done server side and the user never has to download spam. But this would be in the distant future. As is, this should not be an overly taxing addition to the current filtering methods, although not being a programmer I am most certainly over simplifying the case.