Hi!
I have been using Cubase and VEPro for about 4 years.
I think the standard advice is to have a minimum of instances in the server. This means a lot of work to build every instance in a meaningful way. BUT it is my experience that I have more crackles and pops if I use this method than if I use many instances!
Let me give you an example:
I have just recorded Air by Bach with VSL Synchron Strings, using my standard template: 5 midi tracks feeding a rack instrument (Cubase) and 1 instance on my new Slave PC containing all the channels and presets. The audio output in this set-up is routed to ONE bus only, so Cubase receives a stereo pair! This set-up uses around 5% CPU on the slave
Then I tried the much simpler approach, 5 track instruments each with VSL server vst2 and feeding 5 instances on the slave, with 1st violins, 2nd violins, violas, celli and basses. Now audio returns in 5 stereo pairs, each track is a unit - midi and audio on same track, and the server reports 5% CPU use. AND NO crackles!
The file size in Cubase went up from 1,6 mgb to 2,1. Latency is 128 I use Apollo on Thunderbolt 2.
To me the last set-up has a lot of advantages: It's much simpler in Cubase, using disabled tracks is really a time saver, if it doesn't forget the routing ;-) And the mixer in Cubase is much easier to navigate.
Until now many members on VI blog praise this approach, when using Logic, because of the new feature where Logic only loads the sounds, that are present in a project, but this has been possible for a long time in Cubase although you have to do it yourself.
The only thing that I imagine that I will miss in this "one instrument - one instance" approach is the possibillity to route all say Room Mix output to one audio bus in VEPro, but it can be done although resulting in 5 more stereo returns form VEPro to Cubase.
Lastly my advice to you will be to ask your question on the VI- blog. Many experts there, many of them also here I know.
Regards Stig