A new anti spam solution by Abaca aimed at ISPs and large corporations has what they claim is a unique and effective method of reducing errors, particularly false positives (that is marking perfectly good mail as spam in error). What they say is that, in addition to well known detection techniques, their ReceiverNet box looks at the reputation of the recipient to assess the likelihood that the mail is spam. The theory is that if you are promiscuous (or unlucky) with your address and you get a lot of span anyway, then there is a greater chance that this new one coming in is also spam. If, however, you normally receive very little rubbish then the chances are that this slightly suspicious one is ok.
I suspect that there is also a bit of psychology here as well. If you get a lot of spam then you are much less likely to complain if the odd mistake is made. Personally, I would be happier if these bulk filters had a coarse mesh, only trapping the obvious spam and viruses. This would be sufficient for their purposes of reducing network and server load. Leave it to me to fine tune it with a Bayesian algorithm which can learn the sort of mail that I receive and want.