In this particular case libraries don't matter much as the tasks in benchmarks are mostly algorithmical, e.g. like finding spectral norm of matrix, summing numbers or solving n-body problem. Maybe for development of language it would be great if somebody could write some benchmarks to test arc brevity.