TL;DR: RVV does much better than AVX2, in terms of how many loops can actually be vectorized, as well as the size of the resulting code.

And, remarkably, some categories of loops cannot be vectorized in RVV due to a GCC limitation which the development tree does not have.

Therefore, expect the gap to further widen in the future.

Apparently GCC 15 (released back in April) can vectorise the early-exit tests in RVV, bring the results much closer to SVE. A pity they weren't able to update re-run their tests in the four months before publication.