I had this thought today on my way home and I thought I would share it, if only to be able to say later: "I saw this coming years ago" :-) The following may sound technical for the non-geek, I'll explain myself better if you ask questions, but if you're bothered by geek-speak, you probably just want to skip this post.
One of the big problems with online communities is the load on a centralised server. The wider your user base grows, the greater the load on your servers. This issue leads to the second problem: corporate control and need to use advertising as a revenue stream to support the server load.
Now with most sites, this in an unavoidable problem. After all, sites like imdb.com are largely a repository of information in a centralised location for access by all. However, with community-oriented portals like MySpace and Facebook, the content is largely member-generated and hosted in one place either out of tradition or because the user base lacks the capacity to host that information locally.
But as hardware gets cheaper and wireless access more prevalent, we're going to start seeing a lot more in the way of mobile computing. The inevitable result then is that we'll be able to host our own information on our own machines and provide (or not provide) that information to the public through an open protocol.
So, for example, under this system, you would go to a central site (for the sake of argument, we'll use Facebook) and search for "Daniel Quinn". The site will have a basic record that these n people match your search criteria and that their information services can be found at their respective addresses. You could then issue a friend request to that user which would be managed only between you and me and the servers we're running on our personal devices.
Once a relationship is established, it exists only on my device and yours -- the key being the host site doesn't store this information, doesn't need to collate and back it up, or handle the bandwidth requests for other users wondering who my friends are. All relationship information, images, shared notes, etc. etc. would be managed by the devices operated by the user you're querying.
This sort of thing is a while off I would think. Frankly though, I think the real limiters would be the writing of an open and scalable protocol to handle a network like this, and the fact that most computer users are idiots who would have no idea how to configure their own server. A system like this, however inevitable, will have to depend on really slick software that does the majority of the work for you, and a solid, open framework upon which to base each application.
Ok, I'm done. Rant over. Wait for the "I told you so" in about 10years.