By scale it sounds like you mean adding/removing CPUS, RAM, Disk, Network, etc. As it stands the only two of these resources that can easily be modified while the instance is running are disk and network (via Cinder volume attachment and Neutron network attachment respectively).

There is a proposal to implement a "hot" resize function in Nova that would take advantage of the fact that modern hypervisors (and their guests) allowing for adding/removing CPU and RAM while the guest is running but this remains unimplemented.