Adrian Holovaty is looking for information on how to run a state of the art browser engine in a headless fashion. Basically he wants programmatic access to all of the DOM information that you can get at through Firebug (or any decent DOM inspector).
I don’t know what he’s using it for, but you could use such a system to write an awesome testing tool. I suspect the overhead would not be worth it for most projects.
There are already a bunch of good ideas in the comments. Definitely worth reading.