I first read most of the article, and the comments, around 5:15 AM this morning. I finally got to really go through all of it now though.
A4mula, I understand your comments, but they miss what I think is the point of the SBM articles, which is to test and learn things. If someone built this identical machine today, after reading the article, I'd join you in calling them a fool. Beforehand though, I don't think a lot of the downfalls could have been predicted.
I think this was a very interesting build. Among its lessons, SLI with Fermi is infeasible unless the build is planned from square one to include it, and then it becomes merely wasteful. Another is to not count on particular OC-ing results. We're told time and time again that each chip is different; well here's an example of one that didn't do so well. That's another reason though, to focus on stock performance in a value comparison, because only that is a sure thing.
As the article states, this machine was designed to win benchmarks. I think that is how its results should be judged. Personally, I think many of us are more likely to build for usability, I know I am; but that may not be a fair way to judge how this one turned out.
If I win it, the only things I think I'd do is yank one Fermi, put in a SSD, and make the OC a little more conservative. Now THAT machine would do very well on usability, and my non-twitch games would do quite well at 1920x1080 on one 470.