An individual's motivation, amount of dedication to practice, ability to learn, etc make a much bigger impact on the time to learn than generic stats
- Normal distribution.
- You probably wouldn't want to learn to drive a car (in order to get from point A to point B) with all the obsolete stuff, such as the usage of a hand-crank.
stats which really haven't been compiled in any scientific way in the last few decades.
What's "last few decades"? I haven't managed to get hold of Marsan's studies in 70's and 80's, but I've found some remarks, that they were quite decisive and done on a large sample.
And there's actually quite a lot of recent research in cognitive aspects of computer use, including typing, with software like ScriptLog.
Learning touch typing on a new layout while using non-touch typing on the same layout or another one will confused the hands for most. It takes a tremendous amount of discipline to keep two distinctive muscle memories when doing essentially the same task.
The more similar the skills are, the higher is the risk of confusing them. That's the same in music, dancing, martial arts, languages and plenty of other fields.
There are plenty of people fluent in both QWERTY and DSK, for example; sometimes there are triggers involved (environment, tasks, physical keyboards), sometimes there aren't any anymore.