In the original paper I wrote that laid the foundations for what would later become optimistic rollups, I analyzed the possibility of a majority-censorship attack and proposed a mitigation, namely: drivechain incentives.
If miners engage in clearly-visible censorship over an extended period of time, a UASF can be used to coerce them to behave properly. Also note that we have historical evidence that bribing miners to engage in such an attack on a large chain is a practical impossibility.
Finally, Ethereum is inherently resistant to censorship, as enforcing a blacklist without explicit access lists leads to opening up a trivial DoS vector for the blacklister. This DoS vector is most definitely not “nearly-zero cost.”