There are some musing by me on the possible performance implications of serial reuse in the entry for 6-Mar-2000 in the Ixora Q&A page on Memory. The point to realise is that the frame chunks are specific to a particular library object and cannot be shared between different packages. Given that you are using persistent connections, the savings are probably not going to be huge, but I would say that it is worth a try unless you are at risk of library cache latch contention.

I am working on a web app with a pool of persistent database connection, but I have several packages where the state does not need to be preserved. The package is called, performs a task, and nothing else happens thereafter. According to the Oracle docs, the lack of state preservation should be a perfect reason to use the SERIALLY_REUSABLE pragma. This should keep the memory footprint per session minimal.

In general, this should work out well. Most of the time, certain packages will be called only once a session. However, there will be times when a loop may occur and the same package might be called repeatedly. I am wondering how the repetitive calling might affect performance. I can find nothing that talks about this type of scenario. If I am going to take a significant performance hit, I would just assume waste the memory per session. Memory is cheap compared to user satisfaction :)

Any guidance that might help would be greatly appreciated, even if it only helps me run a better benchmark.