So suppose that there are N potential outgoing peers in total, M of which run full-rbf. That means the probability of any one peer not running full-rbf is (1-M/N), so the probability of all 8 peers not being full-rbf is (1-M/N)⁸

There are approximately 5000 IPv4 listening nodes. So for 50% probability of connecting to at least one full-rbf node, you need 415 full-rbf nodes, or 8.3%. For 95% probability, you need 1561 full-rbf nodes, or 31%

Of course, if you are running a listening node third-parties can ensure full-rbf replacements get to you by just connecting to a high % of the entire network. With just a few thousand public listening nodes, that's entirely feasible.

Also, if those odds aren't good enough for you, you can also just run the preferential peering patch: https://github.com/petertodd/bitcoin/tree/full-rbf-v24.0

It advertises a FULLRBF service bit, and ensures that you have at least 4 connections to other full-rbf nodes.