Assigning to `b.N` in benchmarks distorts the results
