This thread got me curious as to whether I'm full of it, so I did a quick Google search and found this experiment (
http://www.humanfactors.lth.se/filea...re/Horizon.pdf), which tested a group of pilots with both types of attitude indicators, and the sky pointer type yielded
5 times as many errors as the GA type. This is despite that the entire test group had more recent experience with the sky pointer type in real life.