Hey,
When you have a situation like described, a server that is getting both physical and virtual metrics, you will have issues with the server's reported configuration. The server will be reported as physical or virtual depending on which data source migrated last. So, in this situation, would not explain 'duplicate metrics', as one is reporting GBL_LS_VIRT_CPU_UTIL, LS means 'Logical Server' or another name for a VM, it is a unique metric that is reporting CPU utilization relative to the VM. So while this is a different metric than the GBL_CPU_UTIL, the values would be similar/same, but NOT stored as the same metric.
In the situation described, you would have entity duplication, meaning you would have a virtual entity and a physical entity.
To avoid metric duplication, ideally the VM's would not be included in the data pull from the CDM probe. This can be accomplished by careful selection of systems in the CDM probe.