As we talk about here: http://developer.rackspace.com/blog/how-we-run-ironic-and-yo..., each instance is provisioned with redundant 10Gbit network links with minimal network over-provisioning in each cab.
oh awesome; did I miss that in the article? Reading is hard...
The one area that this announcement doesn't address, however, is contended networking. Rack local servers with a 2G (or ideally 10G) full port-to-port simultaneous switch did amazing things for our app's performance. When you're looking at whole-app performance, you're sensitive to contention on all of cpu, io, and networking. Hadoop is particularly sensitive to contended i/o, but secondarily sensitive to networking.
edit: s/big data/large data/ -- people who say bigdata are tools