Thank you for your response.
We have been keeping an eye on the GC, and we have used different GC policies to test the results. We didn't still see the consistent results on linux. We have configured more heap size 3G, which is definetely more than required in our case.
I would enable logging and check out the GC behavior in more detail, though we eliminated it as an option initially because we see only 60% of the heap getting utilized during our run. And we do not see any CPU utilization.
Somehow the threads are getting collected in the queue resulting in more time for processing the requests. It becames very incosistent for different runs. Any other pointers will be highly apprecaited.