Of course, the t2.micro instance is fairly weak and you should know more about how Amazon provides this computing power for EC2 instances. You can read more about them at https://aws.amazon.com/ec2/instance-types/.
If you use more powerful machines with a higher number of cores, the performance difference will be more visible between different configurations.
When it comes to results, it's no surprise that the default installation of BLAS and LAPACK gave us the baseline performance and optimized versions, such as OpenBLAS, ATLAS, and Intel MKL, gave better performance.
As you have noted, you haven't changed a single line of code in your Python script and by just linking your NumPy library against different accelerators, you had huge performance gains.
If you will dig deeper into these low-level libraries to understand what are the specific routines and functions...
