Relatively rational people can form deeply irrational groups, and vice versa.

I would probably take a group with rational institutions but irrational members over a group with irrational institutions but rational members.

Of course, rational people will be better on average at building rational groups, so I would still predict a positive correlation in the experiment.