Remember, "MVC" means so many different things. Originally, the "controller" was needed for low-level keyboard processing, something no modern "MVC" framework needs in the same way. (See Martin Fowler's history of the term if you're unfamiliar with it).
So really, all "MVC" means these days, is, "Hey, we put some thought into how to have better separation of concerns in our UI code."
But past that point, it is hard to make simple, true statements that include all of the MVC frameworks out there. Rails controllers are different from Ember's controllers are different from Backbone controllers, etc.
So when someone tells you, "Here's an MVC framework," the only reasonable expectation you can come in with is that they put some thought into separation of concerns. Where that thought led them is likely to be very different from other MVC frameworks.
And in Ember's case, the particular details are, "Our whole philosophy was strongly influenced by Cocoa."
Ultimately, we seem to just now be getting to the point where we can, as an industry, intelligently figure out what concepts do make sense in a single-page javascript framework that "puts some thought into separation of concerns." It's not likely to be any of the approaches we've already tried, because our "javascript platform" was immature, and all we have done so far is borrow ideas (from Rails, or from Cocoa, or wherever).
Another interesting post! It's even more interesting for me, since I've been testing with various JS frameworks in the last two weeks; namely the three major ones - BackboneJS, EmberJS and AngularJS.
Aside from the points you've raised, here's a small conclusion I have made.
"Creating a complex web application with Views, Sub Views & Deep-linking is difficult, no matter what framework you use".
I feel comfortable with BackboneJS because there isn't too much abstraction and convention. Therefore, once you get the hang of it, you know what you are doing (not sure if it sounds right, but it feels... pure, or purer than others).
But BackboneJS does in fact make me repeat a lot - which is why there are a number of plugins and other frameworks that try to cover this (e.g. Backbone Marionette, Backbone layout manager, Backbone.Subview, Backbone.Courier).
After 'giving up' on EmberJS, I've looked at AngularJS. Their example was quite simple and made sense. But as soon as I tried to create a complex app, I got stuck.
Working with these frameworks has been quite interesting, because we are trying to somehow capture the state of the app in the URL (e.g. http://a/#b/c/1/3/2) so that if that URL was bookmarked and user directly goes there, we want the web app to be able to render all the parent + child views, grab the necessary data (sometimes making sure that the order is correct) and display them as expected.
It's all do-able in either BackboneJS, EmberJS and AngularJS, but this type of apps are complex and I'm not sure if any one of these frameworks make it "magic".
I'm going to have to agree with your statement, "Creating a complex web application with Views, Sub Views & Deep-linking is difficult, no matter what framework you use". I tried using emberjs, their docs at the time were all over the place since they were not in-sync with their distributed code-base for learning. I've noticed more people have created good tutorials working with backbonejs and seems the right path moving forward for atleast making something remotely understandable.
the documentation has improved rather drastically as of late. the router api is stable going forward, so it's actually a pretty good time to be experimenting with ember (though this was not true about 4 months ago).
Personally, I really like ember. I think most people would have a better experience with ember if:
(1) they avoided learning off of the master branch during the crazy api changes over the last few months
(2) avoided ember data for the time being (or ever, I have my doubts about how viable it is)
(3) they asked questions on stack. the ember core team is actually pretty snappy at answering questions.
but there is still a bit of lingering weirdness with the router that people should look out for, the things that come to mind are:
a) back button errors
b) awkwardness in the serialize/model hooks depending on whether you're navigating through the app, or trying to get to a url in a new browser window
but despite its shortcomings, I've really grown to love the organization that ember gives to my javascript code base, and how it would make my program scale as it grew in size and complexity.
read-only attributes comes to mind, but that was four months ago and it seems that it's been pretty much worked out here, https://github.com/emberjs/data/pull/303..
Regardless, I still think it's a bit overkill. You'd be surprised to see how much of the basic ember-data functionality you can achieve just by using Ember.A and Ember.Object.
Remember, "MVC" means so many different things. Originally, the "controller" was needed for low-level keyboard processing, something no modern "MVC" framework needs in the same way. (See Martin Fowler's history of the term if you're unfamiliar with it).
So really, all "MVC" means these days, is, "Hey, we put some thought into how to have better separation of concerns in our UI code."
But past that point, it is hard to make simple, true statements that include all of the MVC frameworks out there. Rails controllers are different from Ember's controllers are different from Backbone controllers, etc.
So when someone tells you, "Here's an MVC framework," the only reasonable expectation you can come in with is that they put some thought into separation of concerns. Where that thought led them is likely to be very different from other MVC frameworks.
And in Ember's case, the particular details are, "Our whole philosophy was strongly influenced by Cocoa."
Ultimately, we seem to just now be getting to the point where we can, as an industry, intelligently figure out what concepts do make sense in a single-page javascript framework that "puts some thought into separation of concerns." It's not likely to be any of the approaches we've already tried, because our "javascript platform" was immature, and all we have done so far is borrow ideas (from Rails, or from Cocoa, or wherever).