QE defines a set of clusters to test against (logical/virtual), however these may actually use the same underlying physical cluster. Because the framework is dumb it will delete and recreate when this condition holds, which may have some fun side effects.
We need to divide things up so it is aware of cluster wide things (e.g. DAC) and namespaced things (e.g. Operator). At this point we can spot when the same host (API endpoint) is used across virtual clusters and only do the cluster wide stuff once. Farewell side effects.