Having read through the previous thread, it seems that there are quite diverse motivations and wishes for this project. On the one hand, some want to address that there's no GNU-like web 2.0 like social space, that's ready to join and use, and on the other, some want to build something next generation, something that is federated and web scale, that adheres to the FSF principles: "free as in freedom" or perhaps "your profile, the way you want it".
The issue with moving quickly is that you limit yourself to a particular model, and later on hit a brick wall. But the disadvantage of building something freshly architected means you can wait quite a while before having something usable.
I propose, why not take the best of both worlds. Many successful projects maintain two branches: an older branch that will eventually be depricated, and a next generation branch with all the new design.
I suggest splitting into two teams, one to take the "best of the web", and stitch toghether something useful, a proof of concept, relatively quickly. Working on the elgg codebase, with maybe some xmpp seems logical.
A second group concentrates on the next generation social net, consistent with FSF principles, that is Web Scale (ie using Linked Data), and that put privacy as one of its core goals. It would be great if the design could be formally or informally endorsed by the FSF and W3C/timBL (Hopefully Matt can present the case, if he has time). This team can also work on federation of different nodes (GNU Social should be able to work from a server, a homepage, a mobile, a desktop etc.), with the intention to move over from one system to another.
I think this way we can get something up and running relatively quickly, but be future proofed, for when the next gen stuff is ready to be released.