The arguments about being able to do all kinds of crazy stuff without bloating code size just don't make me happy at all. Should we include an extra build step and production pipeline complexity for web, something that is supposed to have flexibility and speed-to-deployment as its evolutionary strengths? I can't even remember how much time I ended up spending on JUST the silly require boilerplate, which accomplishes mostly nothing because it's not like I ever want to fetch javascript asynchronously. CDN JQuery is a 304-not-modified and my code is gzipped by nginx if the page is over 4kb. Hey look, it's really fast and I didn't work at it.
JQuery 2.0 is about to be out, so goodbye old IE bloat. I'm happy with ever increasing internet speeds, ever faster browsers, and a plethora of small libraries ideally suited for various legs of the journey. I've never used underscore, knockout, or angular, but I know that if I ever run into a problem that requires more architecture, I should go find out what they do. This can be one's approach towards almost all problems in web. Which comes first: a focused javascript library or an implementation of it on top of the Dojo (or any other) API?