We should introduce a simple benchmark that measures how expensive it is to reject requests at the rate limiter. Ideally, this should be as cheap as possible, so it will be useful to track if this changes over time.