Do human-machine combinations still beat the best computer chess players, or have computer players gotten good enough to beat even the best computer-human duos?
Two years ago, top GM Nakamura played a match against one of the top engines, Stockfish. In the first two games, he was assisted by a very strong engine, rating about 200 points lower than Stockfish. Still, this wasn't enough to win, he even lost this part of the match 1,5 - 0,5. More details, including games, can be found here.
Of course, a human assisting a computer player will be stronger than the same computer player alone, but this gives you an idea of how little we (human players) are able to add at the moment.
First and foremost:
The centaur should get at least 50% of the points.
As the centaur is not obliged to intervene at all, he could just let the software do the job, so it's best computer - best computer... with an expected result of .5.
How much better the centaur will perform depends on his ability (not necessarily chess skills alone) and rules. If he's allowed to to change engines and knows about strengths and weaknesses in the programs for certain kind of positions, choosing the right engine for the given type of position might gain an advantage. Also, the human might be better at time management.
Generally speaking I think the better the engines get (compared to the operator), the less advantage they have. And of course it's easy to ruin the game by bad decisions.