Also, it is possible to check on testnet, how hard it is to reorg a single confirmation (especially, if a block is ASIC-confirmed). Sending 9950 tBTC in fees was not reverted, even though the block was CPU-mined. I don't know exactly, which amount would be big enough, to get it reorged, but I am accumulating test coins, to test that kind of scenarios.
Edit: More than that: small pools have too low hashrate, to succeed at such attack. And big mining pools have too much reputation to lose, because if some pool will attack, then by looking at chaintips, node runners could prove, that "the pool named X stole those coins". And then, you can imagine, how many miners will switch to a different pool in that case.
So, full-RBF is applied only to transactions, not to the whole blocks. In case of blocks, it is first-seen. And you need two blocks, to trigger a reorg, a single block is not enough, because first-seen block will be different in 99% of mining pools, so they will extend the original chain.