Until I know that there's really a problem, I don't.

If there is a problem, then measure before making any code changes. Sometimes the optimization is where you think it is, but much of the time it won't be. Then measure again.

- when faced with more than one obvious way to do something, what's the determining factor?

Pick the way that's clearer. Code has multiple audiences: one audience is the computer, another is the person who picks up the code next. I try to make sure things are correct for the former, and clear and concise for the latter. If you write all of your code with the assumption that it might be tacked up on a wall for passers by to comment on, you won't go too wrong.