"Is it the instance administrator's fault for not blocking troll instances/troll accounts?"
Wasn't that how it worked on Usenet? Usenet providers were expected to accept and handle abuse complaints against its users, and if they failed, the community of News admins could declare a UDP (Usenet Death Penalty / i.e. defederation) against that provider.
Relationships between users and providers were more substantial and less easily replaced, so providers could put more force behind
@jerry
it when demanding that their users behaved, and providers could also usually directly tie a Usenet post to a specific, identifiable person -- a particular student or employee, or a particular customer.
Even so, the system eventually failed.
Still, I'm not sure a federated, decenteralised system can work at all without, at some level, the idea that "if you don't make your users play nice, we're not going to play with you at all."