For several reasons, I am unconvinced by the tests linked above on backup devices. First, those were done on backups above the device only, so it's unclear at best whether the results apply to backups below the device. Second, it's unclear whether the subjects were experienced or naive users of rap backups, or even experienced rappelers (cavers rappel, but not nearly as often as climbers) - and people who are inexperienced with belaying or rapping *do* death-grip the rope above the device, while experienced folk leave it alone. Third, one of the dangers they cite is that a backup below the device might get jammed open by the device - which is easily prevented by length adjustments, and a red herring of an argument (of course it will fail if you set it up to fail).
I still think that a backup might have saved Skinner (though if he was going to back anything up, it should have been the belay loop first).
Because of Fish's tests (though they depend on how your leg loops themselves are constructed), I think I'm going to move towards rapping on an extended device, with a backup below and clipped to the main leg loop tie-in point.