There seem to be multiple regressions affecting ip6base, especially on tx2.
See trending graph [0].
One bisect run converged to dpdk upgrade [1] seeing 6% MRR regression, but during the search either other regressions were visible, or the test results are not consistent enough.
Quantify the performance impact of DPDK version change for 2n-tx2 but also for other testbeds, so we have one known effect before investigating other possible regressions.

[0] https://s3-docs.fd.io/csit/master/trending/trending/ip6-2n-tx2-xl710.html#t1c
[1] https://gerrit.fd.io/r/c/vpp/+/34705