One thing that is pretty annoying about GUIs is the fuzzy factor: there are often a few pixels that could be the button, and could be the background, but you can't tell which it is for sure -- so they're effectively off limits. A keyboard doesn't have this issue, so it's a lot easier to turn commands into reflexes -- your reflexes are less likely to be wrong.

This is also why I switched from T9 to whatever the other alphabet-entering system is for cell phones: it was just too frustrating not to know exactly what the output would be if I typed quickly.

I'm amazed, too, how certain words I have "over-learned" to type. For instance, it's almost impossible for me to type the word "create" without adding a 'd' to the end unless I make a concious effort to avoid it. I also typed "amazing" in my first sentence before quickly backing out the "ing" (freudian slip of my ego?).

I wonder how this over-learning might be related to reading behavior. I'm sure we've all seen those sentences where the author transposes the middle characters of each word while leaving the first and last characters the same; the sentence is suprisingly not difficult to read. Maybe in the same way our brain has a "good-enough" pattern matching behavior while reading, it quickly recognizes the pattern of the word we're typing and says, "I know this one, I'll take it from here."

Some graphics applications (like 3D modeling package) have pie-chart like menu system; for example, if you type SPACE, an overlayed menu appears around the mouse cursor and just moving a mouse a bit to that direction and release SPACE triggers that item (you don't even click).

Watching the experienced users using that interface is just amazing. They don't read the menu items; they know which direction is what by heart, probably in muscle memory.