I think this post is a bit premature. Most first iterations are not particularly interested in performance or memory consumption, there just isn’t much value in it. Not to mention: it’s not released yet!
Chill out everyone.
They’ve also implemented some parts in C++, like the spell checking (where on OS X they call to the system spell checker). There’s a good chance that they could move some of the editing code into C++ as well if they needed, especially if they provided a separation between the editor’s data models and how that’s rendered into the DOM.
The interesting question the post made me think was if it is even possible to remove the overhead in their tech choices. I mean it should be clear that they will never achieve the efficiency of vim, for example. There is definitely a floor. But you are right in the sense it is not at that floor yet and maybe that floor is acceptable for their envisioned use cases.