This has been by far the most variable set of reviews I've seen. Usually, with a new chip release, each site (except for a few exceptions) would report similar results (i.e. this chip being xx% faster than a previous chip on average). The HT figures in this case, however, seems to vary significantly. Ace's review (a site I put above all the others as far as credibility) shows very little cases in which HT actually decreased performance. Anand shows a relative percentage graph for HT (very helpful) but failed to show HT/non-HT results in his general suite. [H]'s results were pretty questionable as it was one of the few sites that got nothing but performance gains (and very big ones at that) from enabling HT. Toms, on the other hand, showed a bit of decrease with HT in single applications and an increase in others. While I may not question Tom's numbers, I do question the relevance of some of his benchmarks. Lightwave, Maya and 3dsMax are far more relevant 3d benchmarks than Cinema 4XL. Why didn't he include those? What about games besides Q3A? One thing I did love about Ace's review was the inclusion of WC3 as one of the benchmarks. A benchmark much more relevant to myself and many others I know.
Another thing to add about Tom's numbers. In many of the cases, the "decrease" or "increase" was well within the margin of error of testing. So saying that they are actual "increases" or "decreases" is questionable. Aces has shown some multitasking cases in which HT has significantly accelerated certain applications and other cases in which it has significantly decelerated certain applications (when multitasked with other applications). The cases of decreased performance would be when running both a multithreaded application and another single-threaded application at once. The multithreaded application is usually so CPU-intensive that it takes up both logical processors and even when the single-threaded application is ran in the foreground, it gets "pushed asside" by the other application. Causing a noticable decrease in the performance of the single-threaded applications. In many cases, Ace's review shows very little difference with HT and many times a little increase, although still within the margin of error.