The resolution is so high on the Nikon D3x that the tiniest misalignment between lens mount and sensor and lens results in edge or corner blur. We’re talking 5-10 microns, an almost impossible manufacturing challenge. Why short focal lengths? Because the percentage error for any fixed alignment error is much larger. My experience has been that 50mm on up is the safe zone, with 35mm on down becoming more and more sensitive to alignment errors (that's on full-frame DSLRs).
I discuss misalignment in my Guide to Zeiss ZF/E Lenses, and it’s the main reason for the delay in finalizing my review of the Zeiss ZF 21/2.8 Distagon (see below on D3x replacement). I go into the issue in detail with 4 samples of the Zeiss ZF 21/2.8 Distagon in Guide to Zeiss ZF/ZE lenses. Due to damage (rock impact), I’m now on my 5th sample, though the damaged one is returning to me from the factory, a comparison I intend to pursue. (the foregoing should not be construed as thinking the ZF 21 samples were a problem, the 4 samples were a Zeiss courtesy to investigate the alignment issues— one sample I purchased was definitely “off” however).
The lens mount/sensor alignment issue is not confined to DSLRs, in fact it’s a serious and apparently more severe problem with much more expensive medium format gear. See Joseph Holmes’ piece Problems in the Land of Precision.
Mr. Holmes and I discussed this issue at length, and our theory is that the problem is particularly acute for larger sensors, meaning full-frame DSLRs as well as all medium format gear. While the problem can affect smaller sensors (eg Nikon’s DX or Canon’s 1.6X crop sensors), the problem is less often seen (but see my comments on the Nikon 10-24 below). And the evidence in reviewing the Olympus E-P1 and Panasonic G-1 suggests to me that the four-thirds and micro-4/3 format is much less prone to misalignment.
My supposition is that an angular tilt or swing error between sensor and lens mount propagates into a focus error that is tolerable or negligible on smaller sensors, but grows linearly with distance from the center (fixed angle = greater displacement as distance from the center increases). Thus, manufacturing tolerances come into play, what is tolerable on a small sensor may be visible blur on a larger one. Large sensors suffer here, unless the alignment tolerances are also narrowed. The errors in question need only be 5 microns before MTF (contrast) begins to drop, with errors of 10 microns or more displaying as visible blur. It’s not even clear that the lens mount on DSLRs has that kind of precision.
The net effect is that super-wides need a really good match between lens and camera, something that means (ideally) perfect lens-mount/sensor alignment, and perfect lens alignment. It’s a good bet that at least some “bad sample” complaints are due in part to the camera, not the lens. Look for blur one one side of the frame or other, and you might be seeing a camera problem or a lens problem, or both.
And by the way, don’t stress the lens mount letting a heavy lens dangle from it, like a 70-200VR. Precision gear won’t stay precise when stressed.