Could just limit ourselves to the space of measurable sets. Seems like the natural approach since we're dealing with notions like "most" (almost), "frequent" (probability) here. And it doesn't seem immediately clear why we would need unmeasurable sets for the multiverse.