Paul Graham wrote a bit about spammers' attempts to fool Bayesian filters and some possible counter strategies in “Filters that Fight Back”.

A little thought should make it clear that spam detection is AI-complete. I think Greg Egan (in Permutation City?) even described a world in which people used intelligent agents to filter spam, so spam used intelligent agents to try to figure out if it was being read by a real person before allowing its content to be viewed, leading to an arms race that ended up with some people being concerned about the ethics of forcing artificially intelligent software to do nothing but read spam all day.

It looks like the arms race is heating up. What a terrible waste of everyone's time.