I think this is a terminology issue. There are no slave devices or folders. The “master” feature was originally called “read only”, because writes from other devices are ignored (but it was very confusing in what direction it was supposed to be read only, so the name was changed).
The fact that there are no slave devices is the reason there is no automatic override. Imagine for example that you have a happily working ten device cluster. Then one guy checks the master checkbox on his device, because he doesn’t want changes from the network for a while. With the behavior you propose, any changes made on the other nine devices would now be automatically blown away, without the respective owners necessarily understanding why.
This is not to say that the behavior you expect couldn’t be made to work - it would just require a new kind of device that would explicitly be a “slave”.
That could be a bug… I’ve seen it take quite a while to take effect though, now and then.