I'll stick with my 5D Mk2 - plenty good enough for me. By now, the sensor resolution must have exceeded to resolving power of the lenses: especially the zooms.
This is very true!!! A really good quality lens (non-zoom) would usually resolve to about 3.5 um (or about 285 lp/mm) in the real world
Very roughly this Canon camera will have nearly 9000 pixels across (9000x6000 would give a 3:2 senor of 54MP, this canon is 53MP native), so on a 36mm wide sensor, the pixel resolution on the sensor is 36mm/9000 = 4um 250lp/mm (I have kept to 9000x6000 as it keeps the maths simple!!)
Therefore it is mighty close (if it were monochrome), but you have to take into account that the camera has a Bayer filter on it, which muddies the water a bit as effectively its a 2x2 array placed continuously across the sensor, so the true spacial resolution is reduced, so really if you were taking an image to measure from your sensor would actually be only 8um (125lp/mm), as you would only work from one colour plane (usually the green one). In effect the Bayer translation process 'makes up' - interpolates - individual pixel values based on the pixel intensity of any individual pixel and neighboring pixels, it produces a nice colour image but wouldn't be used in say Industrial Vision for pixel accuracy measurement. When working in colour the Bayer filter really muddies the issue, and depending on the implementation of the Bayer algorithm and anything else Canon may do, the real world resolution sensor is likely to be around 150-200lpmm, as far as optics goes.
So you will probably need L Glass or equivalent to see the sharpest images from this camera, and probably some of the more recent L glass, some of the older zooms may not cut the mustard.
I work in Industrial Machine Vision, and we are forever getting clients who want to measure very small objects, but sometimes there just isn't the optics available to do it (Although there companies that will make custom glass, but the lens are very expensive, and not suitable for a 24/7 industrial environment). We always work in Monochrome for accurate measurement projects, and make use of sub-pixel interpolation where we can on regular objects (circles, straight lines, etc) to improve the repeatability/robustness measurements.
If I wanted to measure accurately in colour, I would use a 3 chip (CCD) colour camera, which has a prism to split the light, and the RGB colour planes are each put onto a single CCD. These cameras are not cheap (not so popular these days), and are more delicate, as the prism location is critical to their working, I have dropped one in the past and had to pay for the prism to be reset!!!
I've hope I've said all that correctly, I've had a very busy week and am very tired, so anything may blurt out.