The total vCPU count returned by the API and as a result displayed on the dashboard, confusingly, does not factor in the vCPU allocation ratio. As a result it only returns the number of "real" CPUs available and it's possible for the used vCPU count to be higher than the total vCPU count.
When you launch virtual machine instances the scheduler will still allow you to use the full number of vCPUs the system is exposing based on your configuration.
See: https://bugs.launchpad.net/horizon/+b...
I found that's a Horizon issue. The cpu ration works well even in Qemu.