World of Warcraft and Wikipedia as Generalizations of the Client/Server Model
A familiar solution to this problem is to transfer some of the load to your users' computers. Download to each of them code that will do whatever part of the jobs is specific to that user and keep on the server the part that has to be centralized. One result is that, in World of Warcraft, if you get disconnected but your computer for some reason doesn't realize it you can continue to "play," moving around a world that is complete with geography and structures--all of those are on your hard drive--but absent of other players and wildlife. This is the familiar client/server model.
It recently occurred to me that a different application of the same principle is central to the success of both World of Warcraft and Wikipedia. Part of what makes the former interesting is that you are interacting with lots of other players--and humans do a much better job of imitating humans than machines do. To put it differently, Blizzard has decentralized to its players most of the job of populating for each player the world he plays in. So as the game grows, so does the number of minds devoted to the job of populating it.
Similarly with Wikipedia. The job of writing it is decentralized to the readers. Any time a new topic appears, it brings with it a new set of authors--the readers interested in and knowldgeable about, that topic. A very powerful application of the client/server model, with human beings as the servers.
One might argue, however, that it is an old application--older than computers. Private property and trade create a decentralized coordination system with the computing delegated to the people being coordinated--essentially the same idea. Double the population and you double the resources to be allocated--and the resources to do the allocation.