Pattern for non-blocking / nonconsuming parallel invites?

Here is another variation of the story: