Oh...that's interesting. That would even be useful on Linux to enable greater resource separation between processes...say being able to lock all Docker processes down to 1-2 cores on a machine and with a hard memory limit they can't exceed.
That's already possible with cgroups. Adding a virtualization layer even on Linux would be wasteful, since Docker runs natively on Linux anyway.
What you mention is a major reason why Linux containers were invented, and is already possible with Docker today. Take a look at the '--mem-limit' or '--cpu-shares' flags for 'docker run', for instance.