Leave HT on....yes you can disable it, but it's not a good idea. My (extensive) experimentation has confirmed that Hyperthreading improves performance in WCG by about 15-20%....not a lot but not bad either
I remember looking pretty closely at the performance difference back in the Pentium 4 days when I ran SETI@Home at college (it was heavily encouraged by the school). It was a pretty solid 30% jump in PPD when HT was enabled. It'll vary from workload to workload, of course, but it seems like it's always about a 20-30% boost for near anything that's multithreaded. I haven't noticed that 20-30% number go up or down much at all in newer processors, but I haven't kept an eye on it too much.
Still though, I don't see why not leave it on.
I've heard people claim that single-threaded performance suffers, but I suspect that's only the case when there's a workload on the second logical core. I'm quite certain that in scenarios where there's only one thread being stressed, the difference is nearly unmeasurable. Modern task schedulers should be aware of which logical cores map to which physical cores and should balance loads equally between cores. (AMD had a similar problem with the FX line until it was fixed with a patch.)
The analogy I like to use is a hamburger eating contest. You eat one burger, reach for a second one, eat that second one, and so on.. The time you spend reaching for that next hamburger, you're not eating. If you used your other hand to grab the next hamburger while you were still eating the first, you can switch to eating the second hamburger while you reach for the third. You minimize the time spent between burgers by making sure you always have the next one ready. Boom. HyperThreading.
The total time you spend holding each hamburger increases (almost doubles, actually), but your total hamburgers eaten per hour increases. Two things at 60% the speed is faster in the long run than one thing at 100%.
If you have two people on your team and the goal is to eat a total of two hamburgers, it wouldn't make sense to put a hamburger in each hand of the first person and have the second person sit completely idle. If the person assigning hamburgers (the OS) doesn't take that into consideration, that wouldn't be an efficient use of resources. I think this is where the single-threaded performance argument stems from.