Right now, all of our benchmarks are done with a completely cold cache. Changes like T90732 should intuitively reduce the load time on a cold cache, but it is not as obvious what it will do when the RL cache is hot. We should benchmark both scenarios so we can measure this.