[yngvizzle]

As wgj says, it's mainly that Python easily interfaces with other languages without having to copy data. This means that we can use a vast amount of highly efficient mathematical libraries written in C, C++ and Fortran.

NumPy and SciPy are the main workhorses of numerical programming with Python, and provide bindings to extremely efficient libraries (mainly a BLAS and LAPACK implementation), that utilizes SIMD parallelism, adaptive thread parallelism based on workload and cache optimal algorithms to minimise communication between the CPU and RAM.

this gives Python multidimensional arrays that have the flexibility of Python types and speed we can only dream of obtaining if we wrote the low-level code ourselves.