Something seems off here - they mention using 100 workers (a worker for every request). I would expect that to perform way more than 40% faster unless there's a

Something seems off here - they mention using 100 workers (a worker for every request). I would expect that to perform way more than 40% faster unless there's a ton of overhead in creating those workers.