The difference is probably negligible, though.
The way I work in Acme, commands (whether built-in or my own) are presented in front of me for execution, so I edit and highlight and execute them. I don't take time out of thinking about code to think about the magical key combinations that do whatever. So it's more muscle memory to commands on the screen and sight recognition than remembering keystrokes. I do recall that many of my Emacs commands were muscle memory, but the more I wanted to use, the harder it was to remember them all. I lost them quickly when I stopped using emacs. Years from now the commands I used to manipulate files in a given directory will still be there in the guide file, waiting for me to run them without error. (I help this along by directory-specific logging of interesting commands.)
I was especially happy to get away from the hundreds of extra options in the GNU tools userland, which made entire other tools obsolete (e.g., sort -u obsoletes uniq, except when you need to uniq -c or uniq -d). All these options are the enemy of clarity in code, so you can more confidently write portable, straightforward scripts, go back to them later and actually read them.
Adding support for moving the cursor with the arrow keys would not involve a very high amount of complexity. I found that Acme required me too much to just let Rob Pike dictate how I should work rather than providing a tool that I can customize to my habits and my needs.
Since we, as programmers, deal with very small targets, using the mouse is not the most efficient way to go.
Some editors allow us to move very quickly to the next opening bracket on the line (Vim's f{ for example) while most editors can't help beyond a certain point (usually the OS's word-by-word jumps). If the only means you have to reach that opening bracket is a dozen presses on the right arrow or the mouse… well, the mouse may have its chances. When you can do f{ the mouse is simply useless.
Also, the motion of the mouse is non-linear: you have typically a fast startup followed by a progressive slowdown until you reach your target. That slowdown happens because the fast startup was done in the general direction of the target and must be adjusted a few times to hit the target accurately.
Another thing to factor is how often you use the mouse in your workflow.
When I do mouse-y tasks (Photoshop, browsing…), the mouse is my primary way too interact with my programs. My left hand is available for hitting modifiers or changing tools while my right hand is almost always on the mouse. In that context, I never loose the pointer and the mouse. Changing from task A to task B with the mouse is easy and costless.
When I do keyboard-y tasks, the keyboard is my primary means of interaction. I can spend a lot of time typing stuff, moving around… without touching the mouse. When I want to use the mouse I usually don't really know exactly where it is — that's a first problem — then, when I grab it, I don't know where the cursor is — second problem — which makes the initial targeting slow and error prone. Once the first steps are dealt with, the limitations in Fitt's law apply.
That's why the mouse is frowned upon among people who use primarily the keyboard.
However, how many people who frowns mouse use knows that you can select a word by double clicking on it, or that you can extend a selection using shift+click, or that nothing scrolls a document as fast as the scrollbar or minimap?
I think that my comment is clear on my position on that "mouse debate": the mouse is an effective interaction device in more than many contexts but there's at least one context where using the mouse is not the most effective thing to do.
Again, the problem is not the mouse itself but the cost of constently switching between it and the keyboard.
In any case we should soon have eyeball tracking to move around. and/or leap motion. also when can I simply shout at the screen and get it to do my bidding?
For years I believed that eschewing the mouse was speeding me up. But I recently had the pleasure of working with a fine gentleman, and a SublimeText user, who is an absolute fiend with the mouse. He has considerably more development experience than I, so it is not an unbiased comparison, but I believe his raw editing speed was a bit faster.
This confirmed my suspicion, originally triggered by reading about the Acme[1][2] editor in Plan9, that such a workflow can be quite efficient.
[1] http://en.wikipedia.org/wiki/Acme_(text_editor) [2] http://www.youtube.com/watch?v=Pydckb9ZU8Y