I have looked at the new P08ic AuAu production that was done with Minuit3 VF to test Anthony's modified VF.

(For details : http://www.star.bnl.gov/HyperNews-star/get/vertex/127/1/1/2/1/1/1/2/1/1/1/1.html )

 Below are the ctb vs. refmult correlations for the HighLumionsity production and for the LowLuminosity one (i.e. for the best vertex VertexIndex==0) .

It looks like the issue of split vertices is almost gone (The tiny band on the left which is basically split vertices , i.e. A high CTB mult but small refmult means a bad vertex with small multiplicity was picked instead of the good one, is a very small percentage (~ 0.1%) of total events) ,shown below on a logX scale:

 And to show the effect of pile-up in the HighLuminosity production:

What is shown is the ctb vs. meanDip correlation for the HighLuminosity production

  -------------------------------------------------------------------------------------------------------------------------------------------------