First is how many physical CPUs are assigned to your VM. If you do have a user that locks up and you only have 1 or 2 physical CPU then yes you are going to experience performance issues. For our purposes, we typically use Dv2 VMs for our smaller deployments and the higher tier Ev3s for bigger deployment. Remember v3 VMs are hyper-threaded so if you are running a 4 vcpu machine then you only have 2 physical CPU so depending on the number of users you could run into issues.
Remote Desktop in general doesn't run that well in low end environments. Anything under 4 cpu will tend to be unstable. Think of it this way, you always want to be able to tolerate 1 cpu being overloaded while other CPUs are available for the kernel, services and the other users.
The second thing to look at is the cleanup of your image. We found that some services such as cortana tend to have high usage when running the multi-session image. The default store app for skype for some reason eats a lot of ram even if it's not used when lots of users are connected. In most of our cases we also decided to go back to using Windows Server as the multi-session preview seems to have a few performance bugs and in any case in general Windows Server is so much more stripped down than W10 that it's just better for performance.
One example that really made us switch was issues with the push notification service having frequent runaways and eating a ton of CPU and RAM.
Since we made the switch to WS we found at least 20% lower CPU usage compared to the Windows 10 image.
I think in general we approach it more for compatibility. If we have a software that doesn't run well in Windows Server then we look at using the Windows 10 Multi-session. Another place where we use W10 is for VDI. Any other scenario, currently, we use Windows Server.
@gab_vision1 Thanks for the notes. I've found D16s seems to be working fell for about 10-12 users logged all using excel and google chrome. I've tried to setup server 2016 but I can't get office to run. It crashes when trying to activate.