yeah, but afaik clients are not using kind `3` to assess relays a user writes to, whereas NIP-65 is specifically for that and I think that's the model that will win.
At the end of the day, it's on clients (or relay proxies and aggregators to make use of these data)