In any case, I think the Canon 5D (more than two years old) is a similar performer at a much lower price, considering the weaker body and the lower ISO base value. This is a lesson for Canon: reportage cameras cannot be designed like studio cameras. The Nikon D3 is a very powerful and versatile reportage camera (social events, journalism, sports, travel...), very well designed for a particular kind of use.
The 5D is not in the same class of noise performance as the 1Dsmk3 or the D3, each of which has about double the photon collection of the 5D with the same exposure, and 1/2 stop less read noise across the board at all ISOs at the pixel level (and a lot more pixels on the 1Dsmk3 to make the "image" read noise even lower).
2. The anti-alias filter is a nightmare when you increase the sampling frequency (number of "pixels" for a given sensor size) beyond some point. Consider this: the 1Ds Mark III has 30% more pixels than the 1Ds Mark II. This is a 15% linear increase in sampling frequency. But I think the real gain in detail resolved is much lower, due to a couple of factors: a) this additional sampling frequency (sensor "resolution") is dedicated to resolve signal from the lens with lower contrast (finer details); b ) the anti-alias filter blurres this finer detail ! The result is much bigger files (in part this is due to the deeper bits tonal separation) and not so much additional detail resolved! From a cost-benefit analysis, does it make sense?
No, it doesn't, but not in the way you suggest. It doesn't make sense because it isn't true. AA filters get scaled in their design so that the radius of their influence is proportional to the pixel pitch. If this ideal is maintained, then the maximum linear resolution of the image (as determined by the sensor, say, as if, it were recording a fine laser beam instead of normal optics) is proportional to the square root of the pixel count. I seriously doubt that Canon would use an AA filter in the 1Dsmk3 that had a wider spread than the mk2, not only in pixels, but in microns as well.
The real issues in viewing resolution on the mk3 are probably due to looking at the pixels themselves, which, with the same lens *should* usually be a little softer, due to the limits of the lens, and this doesn't necessarily go away when you view or print the images from both at the same size, if the downsizing methods used are not true resampling. Even $600 photoshop uses a quick and dirty method of downsizing images that maintains individual original pixel limitations, in terms of noise and resolution. Who knows what it does when it goes to print, too. It may use methods that maintain individual pixel characteristics, too, when it sends the print driver the data in its native color pixel resolution.
And let's face it, Canon's wide lenses are not stellar performers, especially in the corners and wide open, so if you're going for minimum DOF with them, your "in-focus" areas aren't always going to be as sharp as you want. AA filters only "ruin" a very narrow range of optical MTFs; the range that is otherwise just a little sharp, but not completely sharp. The really sharp parts should have the AA filter, and the soft areas, due to optical limits or being OOF are hardly affected by the AA filter at all.
It really makes me sad to see all the current backlash against AA filters. I personally can't fathom why anyone would want to look at or capture an aliased image; they look totally unnatural to me, like an image viewed through fractured glass that just happens to break in simple grid pattern. The sharpness is clearly false, with high-contrast edges in discreet, non-analog, "snap-to-grid" positions.
The ultimate solution to issues like this is higher pixel density; once you get to a certain pixel density, you're going to outresolve the lenses anyway so no AA filter is needed. Noise energy will also be concentrated at frequencies beyond the resolution of the lens, and noise can therefore be discarded in processing without losing subject detail. The issues, of course, with higher pixel density are slower camera operation and computer processing time, and storage space, so growth in those areas are necessary, to make super-high-resolution practical.