OK guys, I'm blown out of the water (or at least my argument is). It seems that inside the Honeywell ADIRU, the RLGs/Accelerometers are NOT arranged in a strict X, Y, Z axis; they are arranged non-orthoganally at obviously known VERY specific angles to one another. (No problem of course with strap down systems). So therefore, the loss of a single RLG/Accelerometer does not necessarlily result in the loss of any axis information, this information is derived from the surviving bits. (Of course it is still vital for a defective unit to be disconnecred from the ADIRU decision making chain; that was the main problem with the QANTAS event in 2005, dealt with an ADIRU Ops Software revision to
3470-HNC-100-03 (version -03).
Honeywell rules OK.