let truss print out timestamps, and the amount of time each system call. I can do that with strace, not sure about truss.
Then, send a request that show the stall problem. from truss output, you can catch some pauses matching the stall period, log in that period is what you should pay attention. When the timestamp, you can always go over the log output and find out that section. If truss output does not pause, it could be something else.