That is an evolving story, which has changed several times before reaching that point.
That said, Angular 2's performance is already looking phenomenal, and assuming some of the stats I have seen out in the wild are to be believed, it can run over 30x as fast as some of the existing frameworks for large datasets, including angular 1. If I had to choose a library to use now, it'd be React, but if I had the opportunity to hold off upgrading, I would absolutely wait for Angular 2.
You can even apply the flux pattern to good effect within the directive while still making use of your existing Angular service infrastructure. Per Evans this would be an Evolving Order over a pair of Bounded Contexts.
Even though the Angular part wasn't very big at the time and the rewrite would not have been too costly, it would not have been wise to move to Angular 2 just to have them pull the same thing again for Angular 3.
As great as Angular 2 seems to be, by breaking compatibility to the point they did, they've essentially created another framework.
For someone looking to upgrade from Angular 1.x, this shifts their thought process from "should I upgrade to Angular 2?" to "which new framework should I upgrade to?".
And for many people, that new framework isn't going to be Angular 2.