For systems with active power management that perform significant CPU stepping, in-browser SunSpider gives results that are surprisingly slower than what one would expect from a command-line run against the pure engine, in the case of particularly fast JavaScript engines on fast hardware. The reason for this is that SunSpider in-browser pauses for 500ms between tests, thus the CPU never properly ramps up to speed when the tests take only a few milliseconds each. It would be better to have a much shorter pause, say 10ms, but then the harness needs to be redesigned to prevent networking and progress indicators from unduly disturbing the results.
This issue was originally reported by Mike Belshe here (his point #3): https://lists.webkit.org/pipermail/webkit-dev/2009-July/008723.html
Created attachment 44828 [details] experimental patch
Created attachment 44833 [details] Patch
Committed r52128: <http://trac.webkit.org/changeset/52128>