Most of the benchmarks are not representative of actual usage. They are mostly large file copies/transfers, extended reads and writes. Actual usage is dominated by small read/writes. That performance may vary greatly what most benchmarks measure.