I can see Harv's point.
We have a lot of Resellers using Cloud VPS, where we can expand the size pretty quickly without downtime, even though they are KVM (disk space wise), a reboot to change the actual tin driving the OpenStack machine.
I suppose our standard unit got so big, due to the Elastic-Sites hosting we do. Where you can have VPS style performance, using CL for isolation. This way you get the performance you need, but we still manage everything for you and you get all the goodies thrown in that shared hosting gets like R1Soft hourly backups restore by plugin etc.
You are right though, once moved away from the monster servers, a "just right" fit is better, otherwise the customer end up paying too much.
Overprovisioning does cost.
From what I've heard here today, looks like CL needs to make good it's promise to etablish limits on OpenVZ, but due to kernel limitations this will have to be OpenVZ7