The post doesn't mention the environment in which you're running Cassandra. Any chance you're running it in the cloud (EC2?), or are you running it on real h/w?
We are running in a virtualized environment, we've done a pretty good job at adding capacity for performance and space reasons ahead of the demand. Loadbalancing is pretty hands off, its just something that takes some time and fundamental understanding, as there are still ways to shoot yourself in the foot with this system.
The labels under the compaction graph show the four servers all running in the same slicehost datacenter (St. Louis-B).