Stuffing data into fake outputs seems to be quite straightforward for me. There are already mature tools to use (since at least 2023). Probably they wouldn't even need own code nor maintenance.

If they (and others) use OP_RETURN instead, the main benefit is for the full nodes who don't need to store that data forever in the UTXO set. And for those using SwiftSync who also don't have to care for these UTXOs.

Win-win for everybody, perhaps?

The only guys that lose are those miners who currently benefit from direct fee payments to them from NFTers to bypass standardness settings ... and I do not even think they lose "that" much, because their main income probably is oversized images stuffed into witnesses, not OP_RETURN.

And the only "possible" danger I see is that as someone mentioned earlier a new fad could emerge with OP_RETURN NFTs. That's why I think it would not be the worst idea to first increase the limit a bit (to 200-500 bytes, enough for L2s, but not enough for NFTs) and then if no collateral effects are detected, in a few months up to a year, lift the limit completely.