We had a debate a couple of weeks ago about instrument failure recognition
and partial panel. Here's another data point:

http://www.dft.gov.uk/stellent/group...ty_023258.hcsp

It's the long-awaited report into the loss of a KAL cargo 747 shortly after
departure from London Stansted just before Christmas 1999. The Captain's AI
was faulty (or possibly the INU used as attitude input to the AI was
faulty). Despite the presence of a First Officer with a serviceable AI in
front of him, a third standby instrument that was apparently operating
normally and a Flight Engineer who prompted the Captain to cross-check the
AIs, the aircraft was lost along with the lives of all 4 aboard.

Julian Scarfe