James Howison has made a workflow diagram comparing metadata management for academic papers with the systems available for music files. I've made a different version of the academic part of this, because there are a lot of gaps that can already be filled in (and I don't need much incentive to play with OmniGraffle) - click for the PDF:
The main point is that you don't need metadata attached to the PDF when you have a fulltext index, a simple filing system and a bibliographic manager (and if you did, it's not too hard to write a script that will fetch the metadata from PubMed and put it into the PDF Info Dictionary).