Some aspects of performance of the TCP adaptor can be effectively tested by measuring an end-to-end latency of a TCP send.
When each send is done after the previous send finished, this gives a measurement of the lowest latency possible under the most ideal conditions. Such benchmark focuses on the speed of the happy path under no contention and the lowest possible load.
When the transfers happen over localhost, impact of most of networking is eliminated and the test measures impact of dispatch code exclusively.
The pass fail condition was previously discussed in comments for https://github.com/apache/qpid-dispatch/pull/1296