Well, every optical pass through any glass lowers the transmission ration (T-Stop), but the D800E despite having no apparent AA filter still has all the optical elements so this isn't an issue. If the AA filter was fully removed there would be a tiny improvement there but probalby not worth it.
One effect of the AA filter is that it can scatter light such that it has less effective coverage on the pixels with more light hitting the gaps between pixels. Microlenses help somewhat here but are not 100% effective.
Another thing an AA filter does is share the incoming light that might hit a particular pixel between neighboring pixels. Due to the color filter this means there will be a small loss of light as some light might be blocked by the color filter. Imagine 3 photosensitive pixel grouped together tightly, each pixel sensitive to red, or green, or blue only. Imagine a ray of red light of the correct wavelength such as if it hit the red pixel that pixel would receive 100% of the photons and the blue-green pixels 0%. That is what happens without an AA filter. With an AA filter the light is scattered and shared between the other 2 neighboring pixels, so the red pixel might only get 80% of the red light, while the blue and green each get 10% of the red light, which is entirely blocked by colour filter. So in this case the AA filter has cost 20% of the photons, which is huge.
The opposite happens if the wrong colour light would entirely hit a single pixel. So imagine some blue light hits the red pixel, without an AA filter the red pixel receives 0% of the photos because it is blocked by the color filter, and there is no scattering so the other 2 neighboring pixels get nothing either. In a sensor with an AA filter using the same figures as before, 80% of the blue light would hit the red pixel and be wasted. However, both the green and blue pixels will each receive 10% of the blue light, the green pixel throws this way again due to the colour filter but the blue pixel will receive 10% of the total photons.
So the AA gains a little in this case, but the total gains are less. If we add the percentages of the 2 cases then the sensor without the AA filter receives 100% + 0% of the photons, the sensor with the AA filter receives 80% + 10%. The difference would be a massive 100% vs 90%, to put things into perspective the ability of a sensor to convert photons into electrons (quantum efficiency) is differentiated by differences of 1%, 1-2% difference will make a measurable change in final noise.
This is a bit of simplification (most light wont be truly a single optimal wavelength for the pixel) and I don't know what % of the light gets blurred by the AA filter, and the complexities of the Bayer filter and associated algorithms to generate a full colour picture may effective things. But the basic premise is correct, if you scatter light and only 1 in 3 of the pixels are sensitive then you will always loose that light.
From this it is also clear that if you shoot black and white then a camera without a Bayer filter will offer massive improvements for low light. But you loose the flexibility to process by colors.
What would be really nice is a modular camera that you can remove the AA filter, IR filter and colour filter at will, putting them back when ever you need, and ideally choosing which of the 3 you want to have in.
