Just floating the idea here...

Maybe it'd be nice to integrate spam recognition into something like [[http://www.projecthoneypot.org/|Project Honeypot]]'s comment spammer recognition (possibly even using that data) or build a similar service for wiki linkspam in cooperation with Project Honeypot.

A security policy would probably do.

An idea would be to make users fill in a captcha when the http:BL reports that the IP address is a comment spammer.