FWIW I suspect TP multi waits in recent versions of Windows are implemented in terms of an undocumented NT object type called "wait completion packets," which I noticed in the list of NTDLL exports (`NtCreateWaitCompletionPacket` and the like). See also: https://devblogs.microsoft.com/oldnewthing/20220406-00/?p=106434
I don't think that's relevant here. The problem isn't the wait threads, it's the callback threads.