With CSIT-1738 reducing stdev of affected tests, the difference between VPP2001 and VPP2005 (using CSIT2005) indicates a regression in VPP.

This only affects tests with l2xc as vswitch, as l2bd keeps high stdev.
This is true for 2n-skx and 2n-clx, but 3n-hsw has non-negligible stdev also with CSIT2005.

On 2n-skx the regression is more than 5%, on 2n-clx it is less.
On 3n-hsw the difference is bigger, but so is stdev, so bisection is less reliable.