When using streaming of chunks as implemented in https://github.com/cortexproject/cortex/pull/3889, average time to evaluate rule groups goes up.