There is a theory that with long lenses, at fast shutter speeds sufficiently high to render image stabilisation unnecessary anyway, that leaving it switched on can sometimes reduce sharpness (when the operating frequency interferes more than it helps). It's extremely difficult to prove because when you're around the hand-holding limit you will always get a few unsharp shots and it's impossible to say exactly what's caused them. The only way to do it is to take hundreds of images around the critical speeds, using lenses of maybe 500mm-plus, with stabilisation on and off, and then assess every image and see if a consistent pattern emerges based on averages. I have never seen this done in any methodical or credible way. I've tried it myself, without any evidence to support the theory, but I've never done that with a really long lens.
On other other hand, it is very easy to prove that image stabilisation is extremely effective at reducing the effects of camera shake at longer shutter speeds below the normal hand-holding threshold, and that is undeniable. Another benefit of in-lens image stabilisation is that it stabilises the image in the viewfinder, making it much easier to track moving subjects with a long lens.