Distribution of 1000's of VMs over 9 nodes leaves 1 node empty

OCP 4.16.30
CNV 4.16.6

Unknown, can reproduce on current cluster.

1. Allocate 9 nodes for VMs
2. Create 1000's of running VMs (increase allowed pod-per-node count as necessary)
3. Stop all VMs and subsequently restart them if result not immediately achieved

1 node receives no VMs

Relatively fair distribution between all 9 nodes

Confirmed the node can run the VMs by cordoning other 8 nodes.
Behaves the same with or without defaultCPUModel set.
Confirmed multiple matching CPU models between the nodes.
On this cluster, the 1 node is always the same node.