I dunno. Here you are basically questioning whether the CIPA waveform is representative of real-world shooting. That could be, and while they do not describe their data acquisition protocols and do not provide detailed results, the comprehensiveness and detail in the spec leads me to believe that they would not base this whole complicated and expensive spec writing effort on sloppy or sparse camera motion data. So, again, I see no reason to buy your theory.i get what youre saying but here is my own explanation based on the discovery channel feature ... the cipa tests i saw (wobbling the camera, moving from side to side in fast jerky fashion, etc), do not account for this micro movements hence the seeming discrepancy.
Re handling subject motion, that is another can of worms that probably can never be quantified because subject motion itself cannot be standardized. (Another issue here, somewhat peripheral to the multi-axis discussion, it that it is possible for manufacturers to "game" the CIPA spec in the same way that VW gamed the emissions testing. Since the CIPA waveforms and test setups are known, it is probably possible for a manufacturer to tune an algorithm set to suit. He could even go so far as to match VW, identifying a CIPA test run based on accelerometer data and selecting the tuned algorithms for use during the testing.)