The cores are divided in time slices so at a given time a logical CPU is always dedicated to a VM. The VM needs to wait until the hypervisor find a free CPU and this becomes more challange when having more VMs which have multibple vCPUs. So more VMs with only 1 vCPU can be run smoother than less VMs with high vCPU count on a Host/Socket.
Regards,
Joerg