It may not be ADIRUs that Failed
"Grunf, I think you are referring to the design criteria of a failure per 10^9 operational hours. You also mention probability. The probability of all three ADIRU's failing at once is more like 10^9*10^9*10^9!"
I know its late but I feel that I need to comment here. What may have failed is the software that runs the voting triplex not the actual ADIRUs themselves. It would be nice to think that software can be written that is 10^9 but its extremely unlikely. perhaps the spikes in speed and angle of attack looked like failed ADIRUs but were actually what was being experienced in extreme turbulence with way outside expectation 100Kt updrafts.