I am under the impression that Cassandra's performance comes from its distribution capabilities.
In fact for most dev I use 3 nodes on my laptop, and most of our "unit" tests are multi-node as well (closer to integration tests by most measures).