IOMeter and 31k buffers

I am seeing that a 31K message size on IOMeter is giving
me less throughput than a 30K buffer. I believe that the
kernel buffers are 60K wide and from what I can see, the
low troughput is due to delayed ACK algorithm. I have
tried to set through registry a TcpAckticks to 1 instead
of default 2, and saw an increase in speed, but that is
unsattisfactory.

There is also a kb article which warns me about the non-
blocking sockets behavior. However, in IOMeter, only
overlapped sockets are used.

Best regards,
Calin