The links you provided are between the 70-200 2.8-L with IS and the non-stabilized 70-200 f/2.8 L lens, but I have one main overriding conern with the reports on the non-stabilized lens, wher they write, "The resolution chart has been revised to take residual spherical aberrations into account". Uh....WTH is that all about? That sounds like fudging the data to me, with the effect of boosting the figures in favor of the non-IS lens by what they call "half a school mark". Hmmm....the 350D also does not have focus micro-adjustment, which had not been invented when the tests you reference were done.
The other problem I have is the testing on the old 8MP sensor of the EOS 350D; they state that the 70-200 2.8L is scratching the limits of that now low-MP sensor's resolution capabilities. Line widths per picture height on 8 MP sensors tended to top out in the 1850 range. As for the lens delivering lower resolution at 200mm focal length--it's QUITE possible that the 70-200 2.8 IS lens has a bit of curvature of field,and/or a bit of undercorrected spherical aberration; either could be preventing it from resolving quite as highly as it does at 70mm and 135mm focal lengths. Lower resolution vs better bokeh perhaps? Or is the lens tested too close? 200mm might be more-optimized for infinity focus; curvature of field often is a problem at close range, but not at infinity.
Also, at what distance did they do their shooting? Almost assuredly, indoors, at a rather close range. What annoys me though is this, "The resolution chart has been revised to take residual spherical aberrations into account" statement; I think that is a ridiculous statement...the figures are what they are. Another thing that makes it odd is that a bit of undercorrected spherical aberration leads to better bokeh,and while it might cost a few lines on a resolution test, having undercorrected spherical aberration will yield a prettier, creamier defocused area, which is something the 70-200 2.8 L-IS is known for--it has excellent bokeh for a tele-zoom.
Bottom line...these Photozone tests bother me. First lens, sample one was a clunker. Now they claim they have a good sample. On the second lens, they have "adjusted" the spherical aberration figures in order to give us the final results? Again, WTH is that!?
The dPreview comparison of the 70-200 f/2.8 L-IS versus the Nikkor 70-200 2.8 VR lens is a good place to look, for a newer test,with results I would tend to trust more than the Photozone folks with their 8MP camera tests. Their conclusion at dPreview is that the Canon 70-200 2.8 L-IS is the better FULL- FRAME zoom lens, but the Nikkor is clearly the better lens on an APS-C sized frame. Again, I am strongly wondering if the same is not true of the 70-200 f/4 compact L-IS model Canon most recently released; could the new f/4 IS model be optimized for the demands of high-density APS-C sensors?
I dunno...I look at the Photozone results with a lot of questions; how many times the focal length were the tests done at? If too close a range, minor focus shift, or curvature of field at 200mm FL would tend to show the lens at a disadvantage; if the lens has undercorrected spherical aberration, especially at longer FL's, the resolution figures will drop, but the bokeh will look prettier, and if the defocused zone is more-creamy behind the focus point, the "perceived" sharpness will be higher,despite lower rez figures when shooting a test chart indoors. A lens with more undercorrected spherical aberration will tend to blur finely-spaced detail, like test chart lines some geek counts with a needle,which will resolve fewer line pairs, but which will be providing better bokeh. Test chart shooting is not the real world, and the Photozone adjustment of the spherical aberration figures in regard to resolution on the 70-200 NON-IS lens have me highly,highly suspicious of both their methodology and their lenswork knowledge. The lack of information and lack of explanation about their reasoning in their tests of the non-IS lens makes me wonder WHY they are "adjusting" resolution results for spherical aberration. Adjusting upward or downward? Again, another "WTH?" moment.
I dunno....I just am not convinced with lens tests done on older, lower-resolution sensors. The exact lens-to-sensor, lens-to-AA filter array makes me leery of tests of high-end gear done by the Photozone people five years ago. Too many unanswered questions, and too simplistic a testing result for my tastes. Some lenses perform better on different sensors; that fact has been well-proven by many lens test authorities, like Bjorn Rorslett. Each camera's sensor and AA filter array can interact a bit differently with each particular lens design. One thing I do note: both lenses have very close center/edge figures on APS-C on the EOS 350D.