Who can explain that DXO find the resolution of the old EF300mmf/4.0LISUSM is better?
DxO can, and they do. Do you define a 3 lp/mm difference as 'better'? DxO states, "Differences below 5lp/mm are usually not noticeable.
" Also, it's not unreasonable that the 300/4 might have slightly higher resolution in the center, and if that's the case, it would have a higher maximal resolution score on DxOMark.
They go on to state, "As noted above, the f/2.8’s resolution is homogenous across the field - rather more than for the f/4L, even though the f/4L has a slightly higher score for resolution.
Why Canon don't react at this point?
Why would they?
Are the FTM curve of CANON a joke?
No, they're not a joke. What they are is theoretical
curves, generated by a computer from a computer model of the lens design, which assumes perfect execution on production and assembly. Further, those theoretical curves top out at 30 lp/mm, which is less than the spatial resolution of currently available dSLR sensors. DxOMark is generating real MTF data from a real lens, with a real camera. Of lens manufacturers, only Zeiss empirically determines lens MTF curves using real lenses (all others use the same type of theoretical curves that Canon uses), and even then it's different, since Zeiss is measuring the lens itself, whereas DxOMark is using a camera to test the lens, so the camera affects the data.