These perf results are sensitive to using different build settings.
There's an implicit assumption here that the tests are run with the
"standard" settings (i.e. what validate does, and the nightly builds).
The perf tests might well fail if your build settings are different.