@StephenOTT, I think I got your idea now.
In case of intermediate message events we can deploy the processes waiting for the same message, and only in runtime the check will be performed, as at the specific moment when the message is being correlated only one recipient must exist.
In case of start event, we can't even deploy the processes. But this is logical, because we will have the ambiguity for sure. We're just able to perform this validation earlier, as we know from the very beginning that more than one process definition will exist, that can be instantiated by this message.