All the store data is kept inside a directory, the hash of which is mapped to your peer ID using IPNS. This enables all store data to be downloaded and authenticated using the peer ID.
Any change to the contents of your directory causes the root hash to change and the IPNS mapping is republished.
Mappings a numbered so when looking up a store via it's ID nodes will always use the most recent mapping they can find.
Republishing the IPNS root breaks the caching that accumulated on the root, but all objects inside the directory that haven't changed retain their caching.