As I have said too many times in too many places to count, of the four pillars of Science 2.0 I originally envisioned - communication, collaboration, publication and participation - collaboration was always going to be the hardest.

Basically, it doesn't exist.   Here, an advocate outlines why he doesn't use one of the available tools, Mendeley, which is not so much science 2.0 as a reference organizer but perhaps has a kernel that could be used in the future.
 Apart from not being able to use Mendeley with our students because we don't have the client installed on our servers (see big burly guy gap), I always find this confusing. So I'm supposed to work mostly by dropping PDFs onto the client? Maybe that's the problem, I hate PDFs, try to avoid them whenever possible. It's not about squirreling away PDFs (I have OS X for that), it's about sharing information.
A real problem with tools, and why I haven't tackled one myself, is that to do true collaboration tool requires a great deal of money, but here an advocate and user of multiple tools doesn't want (or cannot afford - it doesn't matter) even a small fee to use it.   That means the business model has to be free and giving something away for free means a user base homerun before investors can get their money back.  Investors only invest in homeruns, of course, but most people get funded with the expectation they will have self-sustaining revenue and not have to count on an acquisition exit.

One thing that may be attainable in the short term, and I may do if there is interest, is an online tool for paper writing.  Yes, GoogleDocs does it, but we use their stuff here internally and we know its flaws.   As we move into HTML5 this may be more reasonable because we could make an app that syncs easily with the online tool.  We do it here in our revisions feature for articles though it isn't an obvious comparison tool, and that's what would be needed.

CiteULike (who have been around a long time - I use them)
Papers (only on the Mac still, I assume)