From the related Github issue that you linked, comment by @Sukera:
An issue may be thermal throttling for intensive benchmarks, as well as cache pollution for shared cache architectures.
That’s why I said:
trying to run multiple benchmarks in parallel really seems antagonistic to the very concept of benchmarking.
Whenever you’re benchmarking something, you want to have the benchmark be the only significant workload on the system, to prevent measurement error.