Having built a massive app or two in the past (one of them for the Australian Mining Industry collecting sensor reading and fuel transaction data): I have to say that most of poeples’ imagined ideas of performance are usually just plain wrong: You should optimize based on reality, not some constructed idea of reality.
Interestingly, the TodoMVC app is based on building a TODO app. If you asked @tomdale to build an App which is performant on massive lists, I’m pretty sure you’d get a different thing out of him. Sidenote: Personally I’ve never had a TODO list with 2000 items in one list! God forbit I ever do.
If your app has 2000 item lists in it, on the one page, you have a serious UX problem!
“Languages, libraries and frameworks don’t scale. Architectures do” and Ember promotes an amazingly scalable architecture. You can adjust it to fit your needs like a comfortable glove.
I really think we could do with improving the ramp to learning to make it ridiculously easy and that’d give us some amazingly good PR. Perhaps at version 2? 
I was talking about this at Documentation & Testing - #5 by JulianLeviston a day or so ago. Education / documentation improvement would also then mean people would reach for the right thing at the right time, IMHO, because they’d understand when to use what.