It is inevitable that relays will deem it necessary - for whatever reason - to filter much more than just spam - including for widespread viewpoint-based discrimination.  But, this is the beauty of a decentralized protocol.  It only takes 1 relay to circumvent system-wide censorship... even if it's to circumvent spam-blocking (for some reason).  Ultimately, users will also need to filter at the client level whatever undesired content their chosen cluster of relays lets through.  But, this would effectively just be a part of the individualized algorithms we've always envisioned anyway.

In short, to any relays inclined to fight spam:  Go for it.  Experiment.  The sooner Nostr embraces this inevitable future, the better.