Does high Load average cgroups give "wrong" overall load average

Assume you have a system with 2 processors. Now create a cgroup and configure this group to use only 1 processor. Populate it with enough processes to give it a load average of 5 (to prove a point); it is now hopelessly slow. I am assuming that the load average in /proc/loadavg will then also be 5, even though a different user is free to use the other CPU with no wait time. Is this correct? Is there a source I could quote for this?