Thanks Jack. I'll look into it further but I'd have to pick up another card to test low and high cards and their respective differences. Benchmarking games and isolated hardware in games is a very arduous and multi-variable task, sometimes quite impossible to do accurately.

To me, it doesn't seem 1280x1024 is benchmarking the CPU at all. It benches the GPU a lot too. Thus you'll have large score variance from GPU to GPU, which is inline with what you found compared to the 7900GTX Michael used.

Secondly, any games if consistent, the coding and settings for the benchmark consistent and of equal length, even if capped, as we have seen time and time again, should still show the respective difference of CPUs but with lower scores. So say if it's capped sub-62, then you would have QX9770 in the lead at max cap followed by all the rest sequentially if it really should be in the FPS lead. The fact that LC shows it the other way round even when capped and him mentioning max FPS he saw as 64, means either a) one or more of his CPU results were under different conditions/settings for the same game, b) he managed to simulate a CPU bottleneck to achieve far lower scores and a different order, where you didn't (maybe vsync had a part) c) that there were some driver problems experienced from time to time, d) that their was more GPU overhead than CPU being expereinced, e) that scores were fudged deliberately (not implying anything - just the possibilities). This needs to be experimented by a few of us now.

I'll ask MS what he did exactly to clarify.

As for improper reviews;

Those aren't the only anomalies and that's not the only review with them, quite frankly. In fact when I ran 3DMark'06 on my Q6600, it gave me 400 points higher CPU score at stock than two well known reviews out there using the same settings and setup. Usually that much a variance is not possible unless you had problems and inaccuracies with your methodology or there was intentional foul play.

An anomaly to all my testing of 5 Phenoms since November is the IMC speed affect. My results do not agree with those, at all.
Another inaccuracy is the power results in each mode (Green/Yellow/Red). We have conducted more experimentation than reviewers to document this (Me and Achim).
Yet another is the difference between TLB button modes, their effects and handicaps or what exactly they do (no, they don't disable TLB patch but only 1/2 MSR on one core which is also disabled with the patch)...
Even further the effect of the TLB errata patch, my daily results don't match.

The list just continues (oh another is the single threaded FEAR bench).

You have the most similar setup to what MS ran in the LC review apart from his being a much slower GPU, P35 and DDR3 in his testing, but these inaccuracies are a lot to do with the time spent on such reviews. Hence, I hardly bother spending time on them if the result is irrelevant to me or I'd be online forever.

I haven't focused on games as I'm not a reviewer and don't intend to be, but even more applicable so, I don't game anymore. Let me boost my Q6600 and try your bench in a while.