I've to admit that I don't really understand your view.
I do understand what you explain but don't understand (or rather don't fully agree) that this is a must or at least the right approach.
Main different between Windows and Linux/Unix design is the "shared DLL" Windows concept. It looks nice but has a lot of side effects, one being that it is indeed difficult to have really isolated environments. When one application crashes, it often impacts other applications.
In such Windows environments, VM helps, indeed, having isolated deployments. Cost for this however is one Windows server license per machine
Furthermore, nowadays, need for Windows (system) reboot is dramatically reduced. You may see Windows system not having rebooted for months.
In comparison, Unix/Linux world do not suffer from same behaviour and sharing same OS is a rather well isolated way is much easier.
Unless there is an need for different kernel, most of the time you can easily stack applications on one single hardware.
If goal is to have different design, like multi-tiers application, then ask yourself root cause of such design ad you may end-up with deployment on different hardware.
Keep in mind that if your "single OS" machine crashes (hardware crash), then only one system will crash. If you have stacked lot of VMs there, then all will crash. This is not a major issue if you have deployed VM relying on central, highly available storage as same VM could restart on another hardware automatically but if this feature is not deployed, then impact of such hardware crash may be really annoying.
Do not take me the wrong way: I'm not against VM principle. This is very useful but very often misused.
- Hardware cost is low compared to administration overhead due to multiple OS you will have to manage.
- VM can be very efficient in term of high availability assuming you went for such design (BTW this is not what you describe)
- If deploying Windows, this is not really cheaper (again, hardware is not that expensive compared to OS). It has some pros in term of "green IT" but some cons in case of hardware crash.
Last but not least, Zentyla design is much more "all-in-one" box than "central interface for distributed services". If this is your goal, the Zentyal is perhaps not the right solution for you.
All of this being said, I do share your goal that is to be able to get flexibility in term of distributed Zentyal components, thus replication and master/slave. What I try to (slightly) challenge is reason why you push for this. Then these reasons are your
and I perhaps don't understand it the right way
I should refrain myself to always try to understand why people are pushing for VM because it always ends up with this kind of debate