What would the methodology of such research look like?

I don't know, you're the one who said you've seen people claiming they've done this research.

How well do you think a perfect machine would do in a large poker tournament?

Machines are not nearly good enough yet at recognizing facial and verbal clues to do as well as humans in poker. And poker requires relatively little memorization and calculations, and humans do no worse than machines. So a machine (with a camera and microphone) would lose to the best human players right now.

OTOH, if a poker game is conducted over the network, and no information (like speech / video) is available of other players, just the moves they make, then I would expect a well written poker-playing machine to be better than almost all human players (who are vulnerable to biases) and no worse than the best human players.