The issue is that there are more operations we can do with `nat`, more properties we can check, than there are with `nat -> bool`. With `nat -> bool` we can basically check `i` indices and so the behavior of any predicate `(nat -> bool) -> bool` is basically determined by the first `i` indices. And that's finite. This algorithm is basically implicitly finding the indices considered by the supplied predicate and then brute-forcing all those entries.
With `nat` there's no such finite cut off point. It's never the case that it suffices to test a predicate `nat -> bool` on finitely many entries to fully determine it. So we cannot do the same brute force search.