If we start with the assumptions that each gyroscope is seeing exactly the same vibration profile (magnitude, direction, frequency), that each gyroscopes parametric behaviors are identical and the signal chains for each sensor were perfectly identical, then I agree, it is fair to expect identical results. The question is, are those assumptions reasonable? Based on my experience, I think that it would be difficult to support any of these possibilities.
First, the packages themselves might be responding differently to the vibration, due to their mounting approach or other mechanical influences that they present to the entire mechanical system. After working with a number of customers who were experiencing mechanical resonance in their ADIS1636x mounting schemes, we published AN-1045, which offers several options for securing the ADIS1636x devices during highly dynamic inertial conditions.
Second, the frequency response of these sensors rely on resistor and capacitor values that vary by as much as 20%.
Third, there are several parametric attributes in the gyroscopes themselves which vary from part to part. Based on what I have seen so far, this is the least likely cause, but cannot be completely eliminated yet.
Right now, I suspect that we are probably dealing with a modest change in the mechanical response, which is exposing a weakness in the signal processing (under-sampling by a factor of 7x).