p21 in the white paper suggests the speedup is ~2x, for the parts that were actually parallelized. That's more interesting (well, to me) than the rather limited increase in overall frame rate.

You may be right that they are doing the wrong thing, but I think the fact it's doable at all is reasonable evidence that their tool is useful. I'm pretty impressed that somebody who's totally unfamiliar with the code is able to jump in and start parallelizing stuff - particularly if they've never worked on a game before, and so might not have any real idea where would be a good place to start, or how things are likely to work.

Maybe my standards are too low.

(Those screwy HUD textures might be evidence that they managed to get it completely wrong, of course ;) - or maybe it's just something simple.)