I think it’s already limited due to how ASICs are designed, and that is unlikely to change, so you can pretty much count on them being small, but there’s still 0.01% chance someone decides to somehow mine a bigger one.
What would be a worse scenario, somebody invents new type of P2Pool that actually works as well as normal pools and starts using coinbase TXes on it.
Then we suddenly kill it with such an upgrade. That would be bad.
This exists on Monero - https://github.com/SChernykh/p2pool
and coinbase TX’s can get pretty big - Onion Monero Blockchain Explorer
My main motivation for this was feeding coinbase TXs to a contract. With 2026 upgrade, a workaround will be available: it will be possible to generate a compact proof that a coinbase TX is bigger than 10kB - without actually needing to push the whole TX as proof. With this, the coinbase oracle can simply skip the too-big coinbases, which should be extremely rare anyway - but by implementing the skip option such TXs can’t break the contract.