This is a nifty idea. The only problem I see with it is that it makes papers far less self-contained. IMO some of the best-written papers let you understand something of their impact even if you're not too familiar with the field. Taking away the lit review hurts that.

You might argue that a single collaborative lit review would do -- so readers could just print out that plus the paper. But often that lit review section is tailored to your specific goals and audience. To do it well, I think you'd end up having hundreds of such lit reviews for a given topic. In which case, why even bother?

(I didn't mean to sound so harsh, by the way! I do quite like the idea of bringing some of software's collaborative tools over to research.)