A little while ago I briefly mentioned the existence of several web sites which do keystroke-level processing. They do this by making hidden, asynchronous requests to the server using script, processing the results in situ, and dynamically updating the calling page (using the DOM). This provides a smooth, integrated effect, far better than the 3270-style batch processing that old-timers like me were once all-too familiar with.
An article in today’s IBM Developer Journal walks through the use of the HTTPRequest object in great detail, bringing this technique into the mainstream.
The unstated assumption here is that the web server is good at processing a rapid-fire stream of lightweight requests. That’s a whole different topic, and I don’t have any good resources to reference in that area.