Best if everyone makes up their own mind on this based on looking at large prints, but second best is evidence from some of the lens testing sites.
The first error people make is bringing diffraction into discussions like this (particularly along with misuse of the word "limit"). Diffraction is not a function of pixel size - as a property of the lens how could it possibly be so?
Take a look at Photozone.de, where some Canon lenses were tested on 8Mpixel and 15Mpixel cameras. Of course we don't know for sure that the corresponding tests were of the very same lens (probably not in most cases), but if you look at all such cases you should become convinced that whether at f/2.8 or f/11 the MTF50 scores are higher with the 50D - athough they are not higher by as much as square root (15/8) = 1.37. Take the 85/1.8 where at f/8 it is almost certainly a diffraction limited lens (all that means is diffraction is greater than all the other sharpness-reducing aberrations at that f no.). On the 350D it measures 1980 lp/ph in the center, while on the 50D the score is 2372, 1.2 times larger. Now we cannot be sure how much of the shortfall over the theoretical increase is due to the lens, and how much due to the camera (but none is due to diffraction which is exactly the same in these two cases).
Another way to look at the question is to go to dxomark.com, and pick a nice lens. If we take the 85/1.8 again and look at the MTF (measurement tab, resolution, MTF) it is about 10% contrast at 80 lp/mm in the center (on full frame, but here we want to get a lower limit for the lens). That is with a 21Mpixel sensor and its AA filter, so the lens has much higher contrast and resolution (we can only guess by how much). Note the figure changes little for RGB or f no. settings.
So at very minimum this lens resolves 80 lp/mm (and probably way beyond, possibly as much as 200 lp/mm, but I can't prove that). Since the AA filter is imperfect it is best to have slight "oversampling", say 3 blue (6 green) pixels per cycle, or 6x80 pixels per mm. That means that a sensor with (6x80x24)x(6x80x36) will be guaranteed to be getting useful information from the lens, that's 200Mpixel on full frame, or 80Mpixel on APS-C. Don't take the actual number too seriously, the point is that even with this pessimistic assessment of lens performance, there is most definitely scope for getting better results, admittedly in the face of rapidly diminishing returns (increasing the pixel count by a factor of 4 will only increase the resolution by a few tens of percent - but one day it will become cheap enough to allow that in a fast camera).
Of course some people who judge sharpness on "100%" images on a monitor won't believe any of this, but look at large prints and decide yourself.