This would indicate to me that BJL's criterion of twice the pixel pitch is somewhat lax. Resolution begins to fall off when the diffraction spot exceeds the pixel size.
Sorry, I did not make myself clear: I was thinking of a rule of thumb for when the decrease in resolution becomes sufficient to be of practical concern, not the first onset of a decline that is measurable in the lab, but might be of only very small magnitude. Erik's graphs above for the 6 micron pitch A100 sensor show a rather modest decline at f/11, the measurement closest to twice pixel pitch, so I doubt that f-stops lower than about f/12 would show a particularly noticeable fall-off in print sharpness.
In fact Erik's graphs might really show that improving sensor resolution does not move the aperture ratio giving maximum resolution significantly with this particular lens (80-200/2.8 APO?), but simply raises the resolution at all f-stops. Maybe this lens has too strong aberrations at f/8 and below to give a clear answer and a high quality prime is needed for a better experiment. (Resolution decreasing once aperture ratio gets below f/8 is not state of the art these days, even for good zooms designed for 'APS-C' formats.)
With the Zeiss 50/2 and D200, f/4 is the peak, but at what point does the fall-off produce visible effects? By f/22 definitely and by f/16 probably, but at f/8 and f/11 it is not clear to me: what percentage decline is needed to be visibly less sharp? That is why I prefer judgments based on print viewing.
For a perfect aberration free lens, measured resolution will be highest at minimum aperture ratio, no matter how low, falling off at least slightly as soon as one stops down from wide open, even at aperture ratios far less than the pixel spacing. But the point at which the decline is noticeable might come only at distinctly higher aperture ratios. So clearly the effect of lens aberrations is also a factor in where the resolution drop becomes significant: proposed theoretical answers in terms of only pixel spacing and aperture ratios can only be loose approximations, valid for lenses of sufficiently low aberrations.