LU factorization is cheaper than QR, in terms of FLOPS. However, the QR factorization does not need to pivot rows, therefore it may turn out to be faster than LU in the end.
The SUPERLU library uses the BLAS library, but has it's own algorithms for pivoting etc. Probably the SUPERLU library would run faster if the system it was benchmarked with here was larger. Therefore, the results reached here are valid for this system, but can probably not be generalized to much larger systems.