PDF Normalization broken using NumCPU

When plotting a PDF containing a conditional observable, using the ProjWData argument to project out this conditional observable and the NumCPU statement to speed up the plotting, the normalization of the PDF breaks. The factor that the PDF normalization exceeds the data is exactly the number of data entries. The problem arises from the use of the NumCPU argument greater than one.

In the macro attached a RooDecay with a per event decay time resolution is created. Toy is generated. A fit is performed. The data and the PDF are plotted in the end.