It would be nice if pure benchmark papers were a thing. Most of the time system papers get accepted for some new idea. The evaluation section is often biased towards the new idea. Independent benchmarks could fix this.