>Is that because the mined blocks themselves are still valid (i.e. the hashes check out), regardless of the presence of an attacker?
Yes. A censorship attack is just that - it's about mining valid blocks that ignore transactions from a certain party, and/or ignoring blocks from other block producers.
> But then how does it help the attacker in an economical sense?
If an attacker has 60% and censors all other block producers (40%), those other block producers won't be able to earn any mining rewards because teh 60% can continuously orphan the blocks they produce, putting them on non-longest-chain (invalid) forks.
In writing my last reply I forgot that difficulty adjustment only happens every 14 days (I can't remember how long it is with Ethereum's PoW, but let's take Bitcoin's PoW as an example), so it's likely that the attacker wouldn't earn any more mining rewards than an honest miner at first.
When that difficulty adjustment happens though, if the miner had been censoring for a good part of the past 14 days, the network would adjust to "consider their 60% the entire hashrate of the network", thus the attacker would start earning the other 40% of rewards for themselves in addition to the rewards they are supposed to earn.
> Doesn't the rest of the network know that the new 51% blocks are tainted, so to speak?
It becomes a very messy, subjective problem. For every block, you would have to prove whether the block producer is part of the censorship attack or just has an incomplete view of the network, which is an intractable computing problem without some sort of subjective heuristic requiring multiple network viewpoints and trusted parties.
