I think it would be pretty easy to show, similar to this article -
Pre-computer era elite GMs and computer-era elite GMs have similar error rates[1], with the caveat that we presume disagreeing with the computer's top choice to be an error. They mostly agreed with the computer's assessment something like 50-57% of the time.
As far as I know Magnus' games haven't been subjected to this kind of analysis, and I recall 2 pretty bad moves in the recent match with Anand, so I'd be interested to see it. I suspect he may be even slightly above 60%.
Should it be something like high 70%'s though I think it would be time to start a conversation about it. It's how Ivanov was busted: overwhelmingly playing one of the top 2 engine moves, which is basically impossible. [2]
1 - http://en.chessbase.com/post/computers-choose-who-was-the-st...
2 - https://www.youtube.com/watch?v=Jr0J8SPENjM