everyone does not have all the duplicated data. they only have the data that they need – the data requested by a user who happens to be using some instance.
handling defederating is a good point. there could be malicious nodes that would be damaging to the network. i suppose there could be a community-mainted ledger of known malicious nodes (similar to minecraft usernames of known hackers), and the admins of the servers would maintain a blacklist. (obviously you configure that your instance’s blacklist would be automatically synced with this ledger)
the mega community idea could be good. where is this being discussed?