Trending [0] shows a weak signal. L2 tests are not affected, ip6 test show regression earlier (that was #4114), and while the change of averages was significant enough to report, the standard deviation is too high for current bisect to have a chance of converging onto the true cause.

[0] https://csit.fd.io/trending/#eNp1UMsOgyAQ_Bp6adYAip56qPU_GtRtNfFBgD706wtNk9VDDyzLzu4MO87PFq8OhxNTJStKJou-DYGl52O4nsaAnGCdJAjO7yiNaN5K5RpsO2roTQZ5VoNoAH0XX-HU2iGM1kYWeYks7cPvKAkx3ULIfyEa0BY1TQQ1gjy6jcz-P9R1s3pE169IrWEDwpvgB0Gi2fP7xWzQ3ZLdTPIq-ih4kvJEiSTnsa4qVlQhHqbZjl-3Q-66-QXe9npwv9IHY0hrIg