This is a sincere question ...
Could the problem be diffraction?
Here's what I'm thinking.
With slower shutter speeds, most of the light (timewise) that falls on each of the pixels won't have been near the shutter blades, hence the issue doesn't dominate.
At higher shutter speeds, the front and rear curtains of the shutter will fall close to each other, with a spacing of a few mm, or even less than 1mm by my reckoning. Thus the light that falls on the pixels would always have been near a blade. From what I remember, this is getting close to the danger zone.
There will be a point where diffraction must dominate, resulting with vertical smearing. I don't know where that level is, but given the effect of diaphragms in lenses, I suspect the level cannot be that far away.
Does that make any sense, or am I barking up the wrong tree?
You are onto something, and no, you are not barking up the wrong tree (although in my case, these hibernating winter trees aren't so quick to absorb the facts)! Look here to Joseph P. Wisniewski's post:http://forums.dpreview.com/forums/readflat.asp?forum=1018&message=34340728&changemode=1
You'll note that, for the figures given there, the shutter blades are actually closer together than you had predicted, but no matter. What's more troublesome is that these shutter speeds for an idealized camera, and the more practical notation of what distances are equivalent to what apertures (i.e. f/5 and f/12) are not true for any camera that is not exactly like these. You'd have to figure out, in other words, how far apart the aperture blades of a given camera actually get.
And of course I wasted a huge amount of time thinking about a response to a question that wasn't asked. In case anybody was interested for my starting point (by no means definitive) about how IS and diffraction may be related (ignoring the interesting possible complication of microlenses, which smeggy's post mentions in passing):
Just to briefly recap what IS is: IS takes a lens or group of optical lenses in the barrel of the lens and moves it in response to gyroscopic readings in order to slightly shift the image back and forth, with the goal of moving a bundle of rays back to the center (even though the part of the scene those rays originate from is now off-center from the front of the lens).
And diffraction is the result of light encountering obstacles. In the above scenario, a moving IS element would seem, in a simple thought experiment, to bend the rays to a point where they encounter an obstacle - the shutter blades, in the case you mention. However, this would depend on where in relationship to the shutter blades the IS group is.
First, however, forget the shutter blades and consider that if the IS group shifts, the edge of the IS lens group that previously was able to collect light from the image could theoretically be blocked by the physical edge of the lens groups (possibly "optical vignetting") or the barrel in front of the IS group ("mechanical vignetting"), because the IS group is trying to "catch up" with the image that has moved relative to the front of the lens (camera moves = lens moves = image moves), and in our theoretical system it may not stop trying to compensate even when the image has moved outside its range. In reality, IS systems do not try to pursue an image that has vanished from the lens' sight. In both cases, the result would simply be vignetting, which is already a problem in many prime lenses, let alone complex zoom lenses with IS. This poses no problem through the rest of the image.
Assuming that IS optical lens groups would logically come before the shutter blades to prevent shutter blades from adding yet more interference woes to the optical and mechanical vignetting, I believe that the only part of the rays from outside that would encounter an obstacle will do so before (or possibly when) they reach the IS group - so the IS group may be slightly larger than needed in order to reduce this obstacle, and the same may be true of elements in front of it.
After the IS group is aligned, the rays will now continue straight down the lens where you can assume normal diffraction effects will set in (i.e., will be determined by highly restricted aptertures).
Diffraction effects can be controlled by the aperture, and in our equation for exposure (Exposure = Light (assuming constant radiation) * Time) we see that the directly related variable, light, is separate from time altogether. The variable most obviously related to IS or VR or whatever, Time, does allow you to vary the light, however.
As a result, if you are shooting beyond the point where diffraction starts to set in for your camera and lens combination, you may find that IS or VR can be utilized to lengthen the exposure. The result is a shot that's more free of shake and appears to have the same brightness, at the cost of more time taken - and in this case, preserves a small aperture.
It may well be the case that VR, when combined with a camera's tendency to overexpose some scenes to meet its internal "this scene is 18% gray like most others" assumption, and with Auto ISO, is allowing some photographers to negligently shoot at apertures that are diffraction limited without their understanding what they are doing. Alternatively, they could be using shutter speeds that are much too long for the scene brightness. I can't imagine at the moment that any cameras will take a bright scene and automatically throttle it down to f/11 or whatever is obviously diffraction limited, though it seems possible - that would have to be a really bright scene! Most peoples' memorable experiences with their camera's programming obsession with scene brightness will result from the opposite situation - when the scene is too dark and the camera ramps up ISO and opens everything else up as far as possible.