Ghost has a number of in-memory caches to avoid constantly making DB roundtrips. If you make a change on one instance then only that instance that will have fired and responded to events to refresh the necessary caches, so you’ve now got different state across the cluster.
There are rough ideas to allow replacing the in-memory caches with a different data stores such as redis that could be used in a clustered scenario to keep all instances in sync. I can’t speak to any timing relating to that work however, it’s just an idea that’s been mentioned so far rather than anything that’s been properly planned.
If it’s something you need maybe open a topic for cluster support in the Ideas/wishlist category?