Ever since we started [0] testing 2n-zn2 with mellanox card (CX556A NIC), the performance has been capped below 28 Mpps.
At that time, CSIT-1791 was causing a similar cap on other testbeds (e.g. below 39 Mpps on 2n-clx [1]), and that cap was later removed [2] as CSIT-1791 was fixed.
But 2n-zn2 is still capped [3] below 28 Mpps as before, proving this is a separate issue.

The task is to find causes, or at least document symptoms (e.g. in telemetry) to narrow down possibilities of where is the bottleneck.

[0] https://docs.fd.io/csit/rls2106/report/vpp_performance_tests/throughput_speedup_multi_core/l2-2n-zn2-cx556a.html
[1] https://docs.fd.io/csit/rls2202/report/vpp_performance_tests/throughput_speedup_multi_core/l2-2n-clx-cx556a.html
[2] https://docs.fd.io/csit/rls2206/report/vpp_performance_tests/throughput_speedup_multi_core/l2-2n-clx-cx556a.html
[3] https://docs.fd.io/csit/rls2206/report/vpp_performance_tests/throughput_speedup_multi_core/l2-2n-zn2-cx556a.html