> trials against different accounts, where you don't gain this information (that the account does not use that password)
I would think that you do gain this info, the question is whether you record it for later use, which seems possible. But the extra effort to do that is a downside.
The upside is of course that 1000 failed login attempts on 1 account is more likely to trigger alarms than 2 attempts on each of 500 accounts.
If you try against the same account, for each trial you gain a (very small) piece of information (that the account does not use that password) which you can use in later trials, which seems like an advantage over trials against different accounts, where you don't gain this information.
But we also know that there are a significant number of accounts using weak passwords. If you keep trying against the same account, you will try the weak (high probability) passwords first, but if they don't break the account then you will run out of those and have to try low probability ones. But if you try against different accounts, you can keep retesting the high probability passwords. So trying against a different account each time is almost certainly more efficient - as long as you don't care about which account you break.