I'll say that after a year's worth of experience with Ember, we've gotten much better at it, but we've still had to add our own components (e.g., a validation framework) or fork off what components and use + fix them before they're completely production ready. We accepted these tasks as a cost of living on the bleeding edge, but admittedly given the rapid pace that the API has changed in the past year, we've acculumated a bit of necessary tech debt.
W.r.t. Ember Data, that wasn't a consideration for us, as the component didn't exist when we were building Dashboard. That said, we chose to integrate with ED in one of our more recent features, and while I'd be lying if I said there weren't any pain, what we took away were learnings about how our non-ED models related to each other and informed subsequent refactors that made the overall app better.