Description of problem:
Currently, performance is well above benchmark numbers using both customer code and the qpid-latency-test harness. Customer is running with TCP_NODELAY on and with a 10GB network. 

 C++ performance is close to optimal.   Latency for 16kb and above messages is half that of JMS client in same environment

   
Version-Release number of selected component (if applicable):


How reproducible:
always

Steps to Reproduce:
1.  run the attached code or test jms client with qpid-latency-test
2.  run messaging tests used using c++ client
3.  be amazed at the difference
4.  Please contact James Kirkland regarding technical detail, testing scenarios and customer environment
  
Actual results:

Executing latency test with: 1000 messages per process with 1 processes.
Messages: 1000 Max: 8.019ms. Min: 1.266ms. Avg: 2.31973ms.
Messages per second (TPS): 431.085


Expected results:

msg size (kb)	qpid1 (ms)	asf http 84.0 (ms)	asf http optimized (ms)	asf http opt + no serial (ms)
1	1.59	1.79	1.55	1.41
2	1.74	1.86	1.68	1.59
4	1.93	2.17	1.96	1.85
8	2.49	2.91	2.29	2.07
16	4.42	4.47	2.71	2.42
64	9.95	13.61	5.61	3.95
128	13.96	23.15	9.22	6.37
 

Additional info: