Although Canon's MTF charts are pretty accurate of late.
Accurate compared to what?
First off, they are
theoretical MTF curves, calculated by computer algorithms using the design parameters for the lens. They are
not empirically measured using actual production copies of the lens. In one way, that's a good thing, because the theoretical curves ignore QC and copy variation. But they aren't telling you much about real-world performance. FWIW, Nikon's MTF curves are also theoretical, but without knowing the algorithms used to generate them, it's meaningless to compare Canon's vs. Nikon's MTF curves. Zeiss' MTF curves, on the other hand, are real data generated from empirical measurements of actual lenses. So, the Canon curves are useful for comparing one Canon lens to another, and that's about it.
Second, and perhaps more importantly, Canon's theoretical MTF curves are scaled in line pairs/millimeter (lp/mm) - a useful measure for film, but a more appropriate measurement for dSLRs is line widths/picture height (LW/PH) since it takes sensor characteristics into account. The fine resolution information (the thin lines on Canon's charts) represents the theoretical data at 30 lp/mm - when you convert that into dSLR relevance, it's 1440 LW/PH. That value is far lower than the resolution of which modern sensors are capable - the Canon 5DII and Nikon D3X can resolve over 3500 LW/PH.
So, current cameras can outresolve the theoretical curves. What does that mean in practice? When you look at something like the
MTF curves for the 400mm f/2.8L IS II, you see that the MTF takes a hit with an extender, but it appears there's not really that much of a difference between the 1.4x III and the 2x III in terms of their
theoretical effect on performance. But when you look at a
real comparison between those two conditions using a 21 MP sensor, the IQ hit is bigger than those theoretical curves suggest, because the higher resolution of the sensors is exposing a weakness that the theoretical curves don't show.