the nice thing about their model is that it's a pure accelerator play. plug it in and go faster, pull it out, or replace it with something else. in theory, there should be no change to the application/query layer. It should be as simple as entering "create/drop extension vitessedata;"
http://pgeoghegan.blogspot.com/2015/01/abbreviated-keys-expl...
http://pgeoghegan.blogspot.com/2015/04/abbreviated-keys-for-...
I wonder whether and to what extent Vitesse benefits from using this technique, which is fairly well known (e.g. it appears in a 1994 paper by Gray -- Alphasort).
As I go into in the second link there, someone reported an order of magnitude increase in sort performance in one case.
For what it's worth, I doubt that this fully accounts for why Vitesse is faster. I don't know that I'd trust my data to it, but I don't think it's snake oil, even if some of the figures shown are arguably a bit misleading.
As for the trust element, yeah, we know it would take time to earn that in the market place.