Cool idea. The readability bookmarklet (http://lab.arc90.com/experiments/readability/) does a great job at extracting the main document text without having to compare to other files, though it's currently all client-side.
However, Andrew Trusty put a wrapper around my code so that you could use it with any website. It's written in Python and runs on Google AppEngine. Check it out: http://andrewtrusty.appspot.com/readability/
If my host was again acting up, I'd link my port.