Spam zombies from outer space

John Aycock and Nathan Friess from University of Calgary have a new paper on next-generation spam generation. Their big idea is to have spam zombies mine the corpus of e-mails sent and received by the user of the infected machine. It than uses that information to craft spam that emulates that user's style, thus in theory making it harder to filter. I've never seen this exact idea before--though there are obviously similar ones--and its kind of interesting but it's not entirely clear if it works. The authors show that they can extract stylistic features, but that's well known. Unfortunately, they don't benchmark their technique against existing anti-spam measures, so it's not clear whether it's any better than existing techniques for fooling spam filters.