I'm not sure how using 20 threads tests threading performance, but there's probably some issue that Ruby used to have that I'm not aware of.
It is impressive though, that to beat the given Ruby time on my machine in SBCL, I actually had to add optimize declarations. Of course, I have no idea what kind of machine the published number were from, and I'm too lazy to try and install Ruby 1.9 myself, but it seems that the old 'Ruby use -> slowness' implication no longer holds.
 Hold on, the Ruby 1.8 test, which takes 22 seconds in their figures, takes 4.3 on my machine. So that would mean Ruby 1.9 is like super-sonic ultra fast. At least on this benchmark. Which is mostly testing the speed of the sorting routine which I suppose is written in C. So what are we talking about, anyway? I'll shut up now.
That does not mean performance is irrelevant, does it? Heavily used web apps benefit a lot from a fast runtime (it allows you to push the point where you have to split across servers forward quite a bit), and being able to do CPU-intensive stuff in the same environment when you need to instead of having to bust out the C or Java or whatever is very pleasant.