How does Berkeley DB manage its files?

Question

I'm using Berkeley DB (BDB) as a persistent store for a JMS queue. When I consume an entry from the queue the underlying BDB files do not immediately shrink, but do eventually. I'm running into issues with the BDB files taking up lots of space on the file system while retrieval performance degrades.

My entry size varies considerably but it is not uncommon to have 400,000 messages of around 32kb each in the persistent queue.

I'd like to understand how BDB manages the files so that I can throttle the number of entries for file size/retrieval performance. Or so I can rule out BDB as my persistent store mechanism.

I am probably searching for the wrong terms but have not found what I'm looking for in the Oracle documentation or The Berkeley DB Book. I would not be surprised if BDB doesn't want me messing with its internals but I would be surprised if (at least) an overview of how it does handle its internals is not available.