Sooo, what about experimental time/motion studies? Corpus analysis? Error analysis? Learning methodology? (Is psychology a science?)
I don't think those studies say that a layout is definitively "better" than another... At least honest ones.
For example, You can do studies of what kind of motion each layout will use more, but you still need to prove that a given motion is better or worse. You can measure speed, and error-rate. But those are closely linked to training, and thus it's difficult to conclude anything. What do you conclude about person X typing faster and with less error with layout A than person Y on layout B ?
If I'm not mistaken, most typing speed records have been performed on Qwerty layouts, I wouldn't conclude that's the best layout using those metrics... Some of the best typist also think that the "one key = one finger" is a mistake (and I agree *), and I don't think I've ever encountered a layout study that consider this kind of thing.
(* on a similar matter, my accordion teacher has a strict rule: NEVER hit the same key with the same finger twice in a row. Indeed, you press better a key twice in succession if you change your finger. And he trained a couple person who won national championships, so I'll trust him on this anyway. Chromatics accordions and keyboard have enough in common to think that they share some ergonomics tricks)
Those studies ARE interesting, but I don't think they constitute a formal and scientific proof of the superiority of a given layout. There's an interesting thing about Ussain Bolt. His running has been studied: he's doing everything wrong (he's even physiologically unable to be a really fast runner). I find him quite efficient, myself, on the running track.
Take one of the Dvorak principle: alternation as much as possible. I've yet to see a convincing study that proove that it's indeed better than rolls (I don't even think it's possible to prove this, or the opposite). After finding the software Amphetype, I decided to program my own software to check my slowest and fastest digrams/trigrams, and found that indeed I type several rolls (especially on adjacent keys with the strongest fingers, such as 'er' and 'ui') faster than alternating digrams.
There's also something I've seen by using my own corpus and tweeked metrics to design my "ideal" layout: I've used 80% of what I typed in the past ~10 years (which is a huge corpus) and... my top 5 letters aren't even the same as the "top 5 letters" of the languages I speak (Mine is E-T-N-I-R, while, according for example to Wikipedia, N is 6th in both english and french, and R is 7th in french and 9th in english).
Even assuming that you COULD define the best layouts based on letters and digrams statistics, since each person types different things, each person will get a different layout. With exactly the same rules as Dvorak (or programmer-Dvorak, or Bépo, french equivalent), you'll get far different layouts.
There's bad layouts (imagine all vowels on the pinkies) and good layouts. Dvorak and variants, Colemak, Bépo, NEO, Workman, Maltron, Asset, Capewell and many others I don't remember all probably have their advantages and issues. You can pick the one that fits you the best, or adapt it to your needs (some have stronger fingers, other not so much, etc.) I don't think there's a definitive metric to order them, even on a given corpus.