Yes, the technology page calls out that on average it is 3.7X faster with just LLVM and algorithmic improvements. We are trying to push more value into Vitesse X by adding vector processing, and maybe include some of our threading work in the next version.
(ck from vitesse data) We haven't exploited this technique mentioned by Peter. We will when 9.5 comes out, and we are looking at it to see if we can put it into our 9.4 extension so people can benefit from it before 9.5. The technique is sound, which should make it go even faster as plans generated by Postgres seem to be partial towards sorts.
As for the trust element, yeah, we know it would take time to earn that in the market place.