That tool is useful for insight, certainly, especially now that they've added the Critical Path Explorer, but again it's not an accurate representation of what real users experience. For one, these tests appear to be performed from the U.S. and so one European site which I've just tested loads in about 900ms, whereas another U.S.-based site, which is in fact heavier than the European site, loads in less than 150ms. Yet there's no way a DSL user in the U.S. will load the site in 150ms.
The statistics in WMT, however, are based on actual real-user measurements. If only they were updated more often! Per-country statistics would be helpful too.