[Spambayes] bayes evasion

[hank at marinar.com]

I'm starting to notice a trickle of spam coming in with advertisements at
the beginning, and some other type of large message appended to the end.
In one case, the large text at the end was a review of the movie, The
Godfather.  While the Plan For Spam FAQ mentions how random words are
handled, it doesn't mention how these types of emails affect the system
if learned from as Spam.  If learned from, how would the classifier react
when the same review comes into the inbox without the advertisement?
Will properly training the spam and ham as such handle this issue?
Will training from these types of messages "pollute" the database,
and thereby make it less effective?

Thank you for any comments,
hank