> Surely they'd at least notice a decrease in installations of the new "too big to download" version?

You are referring to a hypothetical scenario where A/B testing took place.