Fixing this issue means adding hardware-generic Mun runtime performance tests so that they may be used in a cloud CI such that the cloud provider's inconsistent machines don't lead to inconsistent test results.
See #58 for more details
Consider using benchlib, which uses perf rather than valgrind (which is slow)
perfrather thanvalgrind(which is slow)