There are some thoughts on this approach - putting a "traditional" style pool on top of p2pool's framework. I'm pretty sure the folks at mmpool are attempting this currently. However, as ckolivas correctly points out, just because you're using a decentralized back end, you're still centralizing mining power because miners have to register themselves on *your* pool. If I mine on pool A, then move to pool B, I'd have to re-register and effectively start again from scratch as a new miner on pool B.
Now, if the entire thing was distributed (i.e. register on one pool, and mine on any of them) that would be a different story, since every pool would have knowledge of every miner, just like current p2pool.
Ahh, ok I see what you mean. I would suppose though that it should be doable to keep track of that info as well, especially if there's a database backend that could be utilized to keep track of that info which could be replicated across the nodes when the user is initially setup. But then again, it does appear you can mine anonymously on the nodes by using your miner address, so I could assume that doing it that way would remove that dependency, right?