On a 2500, I'm seeing activity across all four cores, with core 1 being favored by 30-60% more than the other three. I wonder if Tom's use of the word 'optimized' has weight to it...
Looking at diagrams in the article, it would seem that performance does not in fact change by much when going from 2 to 3 to 4 cores. I guess hence the conclusion.
Of course, when one has multiple cores, the processing load will be shared on all of them (the operating system shifts it sequentially from core to core), so no one core is constantly at 100%. Hence there is usually activity on all cores, but the lack of optimization for more than two threads would mean that more than two cores are not often utilized simultaneously.
What I find interesting in the article, is the extent to which Skyrim would seem to be CPU-constrained. If that is true, it would mean that SLI/Crossfire may be less relevant than having a really fast processor, that does not need to have many cores.