benchmark baseline check doesn't respect threshold-tolerances when checking already-acquired baseline which is on disk, against static thresholds .json files.
Apparently it's because the benchmarks are not loaded so the tool falls back to the strict threshold-tolerances.

This makes our benchmarks fail even though we have less than 1% and 100ms of fluctuations.