A new kind of spam is easily leaking past POPfile -- tiny PDF files (e.g. in my current pile, all the PDF's from 20-30kb are spam)
In my experience, normal PDF's are usually much larger.
What if POPfile were able to classify based on type and general size of attached files as a "word type"?
This could help trap PDF spam, GIF-file-spam, etc.
Body content size (or lack thereof) is an associated trigger, of course.