It would make more sense to present an editor and ask the user to make a specific high-level change (such as refactoring a function), let the user do so, and then analyze their input to figure out whether they could have done so more efficiently. For instance, in vim, did they navigate by hitting hjkl or use a more efficient /search, and did they hit delete/x repeatedly or use something like c2w?
If the goal is just muscle memory for beginners, I think instead you might want to have specific actions at a high level with the goal of using particular families of commands. For example, swap the highlighted two words in the following paragraph using only full-word commands. Then it could have a very clear and expected input for you to use as a goal and you could practice doing it.
For instance, suppose you're sitting in the middle of a line, and you want to add a word five words to the right on the next line down. j5wi or /sometext[enter]i are both sensible ways to navigate. )wi might make sense if the insertion point is just after the end of a sentence. jwwwwwi is not horrific (given that visual counting may not be as fast) but probably deserves a warning as a reminder. jllllllllllllllllllllllllli is well worth warning about.
I don't think that training on it would be useful at all.
For textmate, I'd recommend learning just control-command T and the program-level shortcuts (closing many windows, opening the "go to file" dialog box, making a new file, etc).
example ctrl-f in emacs. along with the description "move forward a word" i'd also like to see (perhaps on the side of the screen) an editor move a word ahead. That way there are two reinforcements; the verbal description and the observed action
also many of the shortcuts trigger browser shortcuts. I don't know a workaround, but it's mighty irritating.