Is this a theoretic question?
Why do you need to add a message type in the system B and then bring it back to the original system A.
Usually you have one customizing system. Can't you add the message type in A and get it again to B? This would at least be a consistent strategy.
Even I have not tried, but I think that system A is overwritten if you transport the chances made in system B back to A.
This should be a pretty similar scenario like this: a distribution model was maintained directly in production, then a new compliance policy way implemented and it is no more allowed to do it directly in production, so we need to do it in development and transport it to production.
