Hal, I'm afraid I've failed to understand your argument, probably because I'm not properly versed in Bayesian reasoning. So maybe I should just shut up (though you encouraged us earlier to engage in topics beyond our understanding). But anyway, this sentence jumps out at me:

"Normally in the MWI if we are going to use Bayesian reasoning, we have to discount branches by their probability weighting, or else we are going to get the wrong answer."

What I would ask is: is the quantum suicide case sufficiently "normal"? It seems like it's a profoundly abnormal case. In the normal case, you're going to be around to observe the results, regardless of what happens.

I think the idea behind using (repeated rounds of) the 50/50 experiment to prove QI is that the experiment leverages this "Observer Selection Effect" in a way that other experiments don't.

But as I say, Bayes is currently above my understanding, so I'm kind of stabbing in the dark here.