You might have a chance to check it. IF you run in such situation you should stop new clients from accessing and see whether the connections are all back to pool if the requests are finished.
Other option is to monitor the in-use and max-in-use counter, if you see it growing under a similar load during the time you will have a connection leak.
You can also add debug=true or error=true to the cached-connection manager of the JCA subsystem. this will track whether an application does not close the connection at the end.
A Exception or a log statement (error/debug) will show you where the connection is used in your code.