In general, I avoid coding contests, especially if they focus on speed. This idea, however, caught my attention as something I'd like to both participate in, and obviously, to see the results. I'm a reasonably proficient Emacs user, and my config is (just checked) just shy of 5k loc mark. I would love to see if all the hours I put into writing that code make any difference.
I'll have more comments later, but the first two, slightly related to each other, would be:
1. Are you sure using just one language for the code is a good idea? It's going to instantly turn off large segments of people who would otherwise like to contribute - no matter which language you choose. Further, "the most popular language" may not even be appropriate: there's probably less Java, C#, and C++ programmers who use Emacs compared to the general population. I think the first step here should be a poll to see which languages are actually popular among Emacsers. Once we have the list, we can prepare a couple of codebases using the top 5 entries (for example.) I think I can contribute here - I don't think there'll be a language I'm unfamiliar with on the list. I'm also reasonably confident that, in case the improbable happened, I'll be able to learn that language in a few days. (Assuming there are docs available and so on, the usual disclaimers :))
2. Another consideration: how much time would participants have to prepare? I believe this is crucial. In my case, I have a bunch of "first-class" lang configurations, for the ones I spent the most time with, and "the rest," with configs just barely above bare-bones. If you give me a month, I guarantee I'll have all the possible tools you could use with the lang prepared, and I'll have quite a few hours of practicing their usage. If I had to go for it right now, and the implementation language happened to be outside of ~10-15 languages I use regularly or used extensively in the past - I don't think I'd be up for it.
> Starts a fresh emacs instance
Why a fresh one? Do you want to measure startup time (and later, performance in general)? But then people on the latest and greatest would have an advantage over someone working on a netbook.
> a emacs lisp script
I wonder if this wouldn't be better handled by an external tool. I don't know which, but it might be worth considering.
> logging keypresses/function calls might feel intrusive
You don't need keypresses - commands and functions would be quite enough. To be honest, I'd prefer a short chat with a participant afterward, maybe coupled viewing the replay at 10x the speed. What is valuable, I think, is the thought process behind the chains of commands, and "decompiling" those out of the "assembler" (key presses) of people's minds might prove quite hard.
That's it for now, I'll be happy to talk more about the idea, though I think we should take the conversation off the HN. My email is in my profile :)