I'm glad to report that with @Erik_Eckstein's help we are slowly but surely nearing the end of the quality improvement efforts in Swift Benchmark Suite. The benchmark validation infrastructure is finally in place (PRs #20807 and #20074) to guard the quality of newly added benchmarks.
In PR #20212 we've implemented a workaround that enables us to lower the inner benchmark loop multipliers to achieve short runtimes which minimize the accumulated error, while reporting the runtimes adjusted by a legacy factor — this maintains the continuity of the Apple-internal long term performance tracking.
I'm currently in the middle of mind-numbingly boring process of applying the legacy factor throughout the SBS: PR #20861. Once this is completed, I'll look into enabling multi-process (parallel) measurement with the Benchmark_Driver.