You could get a bunch of qwerty users/untrained typists and then train them to use randomly assigned keyboard layouts. Once training is complete, you have them

You could get a bunch of qwerty users/untrained typists and then train them to use randomly assigned keyboard layouts. Once training is complete, you have them type some long piece of text, and then fill out a questionnaire rating various qualitative points like tiredness. Finally, you do a statistical analysis to determine if layouts with more of Dvorak's purportedly beneficial features are correlated with higher comfort ratings. Of course, those features must be rigorously defined beforehand, and you must properly blind yourself when doing the analysis.