I think the problem when you're doing X vs. native is you're implicitly saying "X vs. the best possible situation." I think it's fine to compare it against vector-less operations, but they should also benchmark against more vanilla native compiler settings.