That's a lot of good information and has been pretty much what I've experienced for a single user with multiple VMs running. But I've also found that multiple simultaneous users creates a much more severe burden for Parallels than the number of VMs running. I'm guessing that, for each user, a new instance of the Parallels application is created, each competing for resources. This results in the console user experiencing very jerky and slow performance.In my experience, this is mostly due to not having enough RAM. 8 GB is plenty for most common desktop applications, but if you need to run several VMs at once, you may find yourself needing to assign that much memory to each one of them. If you're running (for example) four VMs, each one with a comfortable desktop environment, then those VMs are probably going to require a total 32 GB of RAM (plus memory for the host environment). If the host doesn't have that much, you will start swapping, which will kill the performance of them all.
Available CPU cycles can also be a bottleneck, but only if the total number of virtual CPUs (across all your VMs) exceeds the number of hardware threads provided by the host, and even then only if the software running in the guest operating system is keeping them all busy. Most of the time, you can overcommit virtual CPUs by a factor of 2:1 or even 3:1, depending on how CPU-intensive the workload is at any given time.
This is why computers sold as dedicated VM hosts tend to have dozens of CPU cores/threads and hundreds of GB of RAM - to avoid these bottlenecks.
It's likely that Fusion handles this situation much better, given its broader use in that kind of environment.