Ideally you'd want to constrain the system to something that is infinitely stiff. I'm not exactly sold on the 3 bolt idea. I'd definitely want one side completely bolted down but I'd support it somewhere in the middle or 3/4 of the way down so when you apply a torque on the other side with a lever arm, you're not also making it a cantilever beam (which I guess the 3 bolt method would help prevent).

And the purpose of the test is "which frame has the best torsional rigidity relative to the others," correct? So if that's the case, you don't need exact numbers so long as all of them are tested and constrained the same on the same rig. Whichever resists the torque to most obviously has the most torsional rigidity.