This affects only CPS tests, larger scales are affected more, UDP is affected more than UDP.
The worst case in comparison tables is 64b-2t1c-avf-ethip4udp-nat44ed-h65536-p63-s4128768-cps [0] at -19%.

The regression happened in this VPP change [1], but the developer says the change is improving thread safety, so some performance decrease is to be expected (although there may be possibilities for better optimizations in the future).

[0] https://s3-docs.fd.io/csit/rls2202/report/_static/vpp/performance-changes-2n-skx-2t1c-pdr.txt
[1] https://gerrit.fd.io/r/c/vpp/+/34077/5#message-88723e1f751eb8cf7eb512bc8811f822efebe66a