It follows from Turing that you can not make a virtual memory system that never page faults (given non-unlimited physical memory, of course), partly because you can't fully reason about execution.

Neither can you reasonably make the assumption that "process faults" (new word?) won't occur in the system that you are describing. And I'll tell you I rather have a page fault then a process fault.