All per account. If account 1 marks a message (sender) as spam, account 2 is not affected. I am blocking at least one otherwise legitimate company that is incapable of unsubscribing me from the marketing mails (despite talking to their helpdesk). I block them now. But other accounts may want to still see their emails.

If you want to learn who the real bad spammers are, I think you need larger volumes of mail than a self-hoster will gather.

The network effect that mox can use is through a RBL/DNSBL, but then an external DNSBL is doing the learning/analysis and mox just uses it.