The key to getting ExtremeProgramming to work in a multi-site environment seems to be system-level tests, perhaps using FrameworkForIntegratedTest.
See also ExtremeProgrammingBoundaryConditions, ExtremeProgrammingMayScaleUp, HundredPersonProject, LargeExtremeProgramming, ScalingExtremeProgramming, XpMayNotScale
Whoa, that's a long list... anyone else smell a refactoring opportunity here?