If not enough node operators upgrade to have all transactions be RBF, the double spend transactions may not sufficiently propagate throughout the network for the miners to see, however, well under 50% would be necessary for miners to reasonably expect to receive these transactions.
Assuming that the average node defaults to 8 connections, then with only 10% of nodes running full RBF then each node has a 57.0% chance of connecting to at least one full RBF node. If 20% of nodes run full RBF, then that becomes a 83.2% chance. It certainly doesn't take a majority to run full RBF for miners to be relatively certain they will see any replacement transactions.
With that being said, it makes little financial sense for the miners to not be on board with this change. All else being equal, it will result in higher transaction fees for the miners.
When considering an individual block, then yes, replacing lower fee transactions for higher fee transactions makes financial sense and will maximize miner revenue. In the long term, however, full RBF will likely serve to reduce the overall fees paid by the community on the whole. If there is no concern about a transaction becoming "stuck", because every transaction can be replaced, then we will likely see fewer people broadcasting transactions at very high fees. They can simply instead broadcast at a much lower fee, confident in the knowledge they can bump the fee at any time if needed. Further, most business which accept zero confirmation transactions only do so if those transactions pay a premium fee, one which is much higher than is necessary at the time. If full RBF ends the practice of zero confirmation transactions, then such high fee paying transactions will also disappear.