Yuri noted in the AuAu data, in the primary vertex error vs. number of tracks distributions there is a second band sitting at higher values.
 https://drupal.star.bnl.gov/STAR/system/files/Vertex%20Error.pptx.pdf

This actually should be due to the feature that st_physics data is mixed by triggers w/ and w/o HFT read-out. See this run as an example
https://online.star.bnl.gov/RTS/2016/triggerApp.php?screen=trgInfo&run=17125001
Most of the triggers (MB, central, BHT1 etc.) require HFT readout, while there are some triggers don't require HFT readout, e.g. BHT3.

I looked at the distributions with different triggers for run17125001. PicoDst data, 20 st_physics files. This clearly shows the high value band in the vertex error distribution is coming from those events without HFT readout.

All events	vpdmb-5-p-sst trigger	vpdmb-5-p-sst trigger && |Vz|<6 && fabs(vz-vzVpd)<3
BHT3 trigger	BHT3 trigger only (only satisfying this trigger)	All triggers with no PXL hits in the event