The addition of the ability to just use querySelectorAll as the selector engine and not ship Sizzle is a pretty big deal, and not something I remember being included in earlier discussions about jQuery 2. qSA still includes a pretty broad set of selectors, and the ability to get the build down to 10k minified and gzipped puts it squarely in Zepto's territory, size-wise, eliminating the need for that particular wheel reinvention.
If you use really simple class and id selectors, the qSA-based selector engine may be fine. However, I think a lot of devs would be surprised about how much Sizzle is doing for them.
qSA does a lot more than class and id selectors; those have been trivial since long before qSA came along. You can also do selection on element attributes, parent-child selections (include strict child selection with ">"), and a few other higher-order selector combinations.
Still, I don't mean to slight Sizzle, and certainly won't stop using it. It's just nice to see jQuery compete with Zepto, since apparently there's a demand for a build that small, and jQuery is better-tested and less brittle.
I didn't mean to say that's _all_ you can do, but people often expect more than qSA can deliver, especially if they've been using jQuery/Sizzle for a while. Sizzle also works around bugs in qSA implementations when they exist.
qSA's design shortcomings have been well documented, John Resig did a blog long ago about them [1].
The list of what you lose if you remove Sizzle is in the native selector file [2].
