No one is blaming the tests. Here is what we know:
1. DBCP is used in most of our unit tests, but only fails in the performFind tests.
2. Different find tests fail, but a failure is guaranteed in one of them (on my machine), and a failure is repeatable (on your machine).
3. One thing that is common to all failing tests is the use of EntityListIterator. I started looking into it to see if a code path in ELI can prematurely close the connection, but I ran out of time.
Right now my gut feeling is that this is not a problem in DBCP, but in our code.