I don't know whether it's clearer than the original paper, but there is another implementation of the technique (or a variation on it) which I spotted in this year's Siggraph material.

http://igm.univ-mlv.fr/~biri/mlaa-gpu/

As to whether or not you'd get a better answer over at the main site, I don't know - I'd say the majority of people interested in this technique are probably game devs...