Even Emacs is smaller, and it's just as if not more customisable/featureful.
Here is a nice rebuttal to that "premature optimisation" myth: http://ubiquity.acm.org/article.cfm?id=1513451
From personal experience, by the time a system gets too big/slow, the inefficiency is usually distributed throughout so that it becomes more difficult to single out one component that's causing it, and there is no easy way to optimise such a system short of redesigning and rewriting it - it can be said that the problem is because "the whole is bigger than the sum of its parts."
Startup speed matters, sometimes a lot, sometimes not so much. It is both interesting because optimizations are a common problem in our field and the different approaches that can be taken are interesting to discuss, but it is also interesting taking a look at how this particular project cares about startup speed and what they have decided to prioritize over it (the social / project management aspect).
With that in mind, if they didn't have those optimizations, Atom would probably be an order of magnitude slower to start up than it is today. It's all a matter of non-functional requirements and priorities, I guess.
On the other hand: as the saying goes, "make it work, make it pretty, make it fast" - which can be interpreted in various ways, like "working software is more important than fast software" (aka avoid premature optimization), and "a well-designed application is easy to optimize".
OT: I've not actually heard that stated before. I've seen "make it work, make it work correctly, (if you have the time) make it work fast"
Maybe I shouldn't be allowed near front-end stuff ;)