i'm not that big of a benchmarker or anything, so i haven't _personally_ done any real tests with hard numbers. (i worked all that out of my system back in the day with a whole lot of extreme overclocking, and one too many peltier-fried processors, giant video card fans and water-cooled rigs)
i have booted with and without smp, and neither one really feels any different (from a responsiveness/interactivity perspective). then again, it's not like using the command line or surfing the net is really demanding. the only time i do notice it is with certain combinations of workloads that balance themselves out well with multiprocessing. for example, i always have seti@home running in the background with only one process, so it only takes up one cpu. if i do something else that would also only take one thread, like make with -j1 or a basic cd record or something, it is handled by the other logical cpu, and there isn't very much of a slowdown in responsiveness. well - not as much as there would be with just one logical processor.
however, how well hyperthreading works is largely dependent on the os and programs that are run on it. hyperthreading came out relatively recently, and the 2.4 scheduler is not particularly good at handling it. however, a lot of development for hyperthreading by people like Robert Love and Ingo Molnar has gone into the 2.5 series.
i would do some tests with the dev kernel i've been messing with, but i still haven't gotten X working right...
but that is an advantage that linux has. the kernel code is changing all the time, which means new fixes and optimizations can be added, while the windows kernel has remained largely unchanged since NT. don't expect to see any really good HT performance from windows anytime soon.
however, there are people who have done real tests with hard numbers, if that's what you want. you might find this ibm developerworks article interesting:http://www-106.ibm.com/developerwork...library/l-htl/
and i seem to remember a post on the LKML from someone from the OSDL who tested and found a 30% speedup on 2.4 kernels, and a 60% speedup in 2.5 (or something like that). i'm sure you could find even more resources on google or something.
'hyperthreading' is an intel buzzword for generating hype. but the technology behind it, SMT, is very real and definitely legitimate.
so is it worth it? now... kind of, if you have the money to spare. it can't hurt and is still better than non-HT. but in the future? most definitely. i mean, 60% performance boost with only room for improvement? holy crap!