To sum up, after phone conference we agreed that:

* with increased TPUT there is no problem.
* Sumit, should have a look at CIFS on read side and analyze the
case where 2 CIFS commands arrive in a single TCP segment.
* we should continue with the investigation in kernel. On the kernel side
we are pretty sure that the resets on client appear because the client
closes the socket and there is still data to be read.

thanks,
Daniel.