From Fitt's law description on Wikipedia [1]:

"It describes untrained movements, not movements that are executed after months or years of practice"

That was exactly my intuition looking at those graphs. There's no way I'm going to be as fast reaching an icon in the upper right corner of the screen after heaving clicked start as I am reaching the top item in the start menu, after some training.

Furthermore, if you don't consider a trained state, this model seems simplistic to me. If you account for the ability to actually find an entry, top items in the list should be faster to click than middle ones - and I doubt a 2D grid is going to be as fast to find something on, until the training effect kicks in.

[1]: http://en.wikipedia.org/wiki/Fittss_law

How often do you target the top item in the start menu (note that it would be the most frequently used) and how often would you target the one in the upper right in the screen (where at least I wouldn't put something I use often – the lower left seems more appropriate for that)?

Well I'm not on Windows right now, but I guess it depends on the use case. For normal use of the start menu almost never. However pressing start key + typing the application name shows the best hit at the top as far as I remember. How is this solved in Win8? Does typing an application name bring up a search result list similar to spotlight on OSX/iOS?

I think if you type the application name the expected behaviour would be to just hit Return to launch it.

But you are right, as far as I remember the search results take over the screen and appear from top to bottom. But I fear you cannot really cater for people who click a button, then start typing and then click again – all of which would be done on the keyboard instead of constantly switching.

Or maybe they need it for gathering thoughts. The time required to switch input devices may come close to the time required to formulate in one's head what is it actually what one is looking for. But that is pretty much speculation ;-)