If performance is the primary concern you wouldn't pick such small server/instance types to benchmark. It's like organizing a drag race between a scooter, a moped and a kid on a bicycle.

Another major flaw is taking results for a single instance type and implying that those apply to all instance sizes and each provider as a whole.

If you're going to do a benchmark at least pick something realistic like the m3.* types: http://aws.amazon.com/ec2/instance-types/instance-details/

At least the author had enough sense not to do the bench on a t1.micro