Knockout is pretty minimal and fast, Ember has not performed very well in update intensive benchmarks. From our testing so far, Derby is somewhere in the middle if you bind everything, but it has granular control of bindings so you can get it to perform about as fast as Knockout.
This is just one small part of synchronizing clients, and Derby takes care of realtime data syncing as well. Synching data with ember requires a lot of manual server implementation. In addition, Ember and Knockout don't have any way of supporting server-side rendering, while Derby automatically renders everything in both the client and on the server.
You are right that you can't partially implement OT. The way that we handle it is by keeping OT operations on defined paths so that they are separately namespaced from other kinds of updates.
I'm comparing details on deferred updates and ease of integrating an OT-type collaboration layer. Derby's already starting to head this way and appears very promising, but I think there are subtle issues both in ShareJS and Derby's implementation. I'm curious how collaboration would work on some of these other libraries, and Angular seems to be thoughtfully architected, if somewhat large.
Personally I'm waiting for 1.0.0 and redone documentation to come out, and then I'll dive into it.
Also, the concurrency models of all the JS frameworks of this nature appear to be pretty immature at first glance. Derby is starting to attach parts of ShareJS, but I don't see much else that looks particularly promising in other frameworks. I'm not sure that you can "partially" implement OT (or mix-and-match parts of different solutions), which perplexes me about some of the code I've seen in the area, even in ShareJS.