Apparently, my question was not enough clear so I will retry to explain :-) :
On VSAN 6.7U1, It is possible the configure some VM on the secondary site with a VSAN policy: "site disaster tolerance - None - keep data on non-preffered site" and "failure to tolerate - 1 failure raid-5" + DRS affinity to enforce the execution of these VM on the secondary site.
In case of split-brain, the stretched VM, with "site disaster tolerance - dual site mirroring" will reboot on the preffered site becaue of witness and number of vote etc...
But what about the VM with "keep data on non-preffered site"? Did they stay up and running on the secondary site? It should be the case because the majority of vote is OK but not 100% sure
Your original description was clear enough, though takes time to set up stretched-clusters in labs for testing just using HOL etc. (I tested it twice with HA and without - split a 3+3+1 into 4+3 using ignore vCenter membership updates and then removed each other site nodes from each sites lists).
VMs with just local-protection on 'Non-Preferred' won't ever even attempt to restart on the Preferred site as the data is not accessible from there.
Outcome is seemingly the same with HA enabled (and aggressive restart everything etc.) and when it is not enabled.
Thanks a lot for your reply. :-)
Have a nice day