Hi Jack,
I think that you can consider those as the same advantages/drawbacks of having in-process and out-of-process data management strategies.
Basically, you will be sharing the JVM memory between your own application objects and you cache entries, which may or may not be a good approach, depending on the use case and amount of data.
For example, running multiple web applications on an application server and having many of them storing a lot of cache entries on the same JVM can lead to out of memory errors if that's not properly sized.
An alternative for such cases would be to have each web app acting as a client and having them access stand alone cache nodes (each with their own JVM) that can hold only data (being a standalone cache server), letting the client decide if he want's to keep LRU - Least Recently Used - entries, for example.
Although there is nothing wrong on having embedded servers on such applications or have them working as peers, always take in consideration that, as said, memory will be shared by your own objects and cache entries. An immediate advantage of this approach is that data access in these cases are very fast.