There's an old saying that you need to build three apps on your framework. That first app shows that your framework works for something. The second app shows that its more general than just the first app. The third app shows that you weren't lucky.
Curious, how was Facebook designed? Are there any post-mortem docs on it? I know that MySpace seemed really ad hoc. I wouldn't be surprised if Facebook was more ad hoc than we imagine it to have been
If I had to guess I'd say that both MySpace and Facebook were completely ad hoc, but the difference was probably that Facebook kept their technical debt down so that they never accumulated an area of code that was too hairy to refactor or replace. MySpace on the other hand probably got complacent in their success and didn't worry too much as their architecture calcified into an unmaintainable mess. By the time they saw Facebook nipping at their heels their code base was past the point of no return, and the company culture made it impossible to hire the level of technical talent they would have needed to right the ship.
Of course I'm completely talking out my ass here, but that's the impression given by the respective product histories.
Maybe just me. I don't know much about the history of the project, except Zuck started it in college.
As the OP points out, protocol cannot be truly decoupled from architecture, especially when it comes to things like routing. There's a huge difference between a Twitter-like @user, an SMTP-like user@domain, a URL-based system like OpenID, etc.
For instance, building around events ("posting a status update", "sending a friend request", "verifying an identity") and abstracting out those events into hooks is how Appleseed does it.
I have no misconceptions that this will hold up to every possible social protocol, past, present and future, but thinking about the protocol as an abstraction, and building the platform out from there, has already shown itself to be very useful.