The architecture was designed so that we perform calculations in groups for efficiency and also dependency, i.e. output1 is an input to a different analysis.
While it is true that in earlier versions an expensive analysis can impact the performance of the entire group thereby negatively affecting other analyses. In more recent releases, we have made some refinement to improve this, minimizing the effect of one expensive analysis. Having said that, if the system is at the limits of scalability, then it may be necessary to have multiple systems.
Can you provide more information on whether you are able to try our latest version to see if your system performance has improved?
As far as I see, this is not about performance but about managebility. If you want to monitor your analysis and the load they produce (you can do this in System Explorer now), you need to have some information about the analysis (Database and Analysis name at least). As all analysis without a template are grouped together, you will never find single analysis out of this group if the group performance badly. Therefore we tell our users to always create a template for any Element/Analysis, leading to a lot of Templates not actually needed as there is only one element instanciated. It is just additional, manual effort in templatizing and working with elements without any benefit besides having a workaround for the lack of information on non-template analysis.