This is something that will be hard to solve. You can't really effectively discern between a large instance with a lot of users, and instance with lot of fake users that's making them look like real users. Any kind of protection I can think of, for example based on the activity of the users, can be simply faked by the bot server.

The only solution I see is to just publish the vote% or vote counts per instance, since that's what the local server knows, and let us personally ban instances we don't recognize or care about, so their votes won't count in our feed.