Jeans can be tough - especially if they are somewhat reflective. For DvSLR video, I recommend avoiding them - not just for keying, but for general color correction. And they often have a thread pattern that could cause aliasing.

And I think there is a misconception about the codec thing. The key isn't to transcode. That can add new artifacts. The key is to work with a bit-deep, 4:4:4 color space, if possible. It really depends on the architecture of the video chain that you are using. If decoding 8-bit, 4:2:0 h.264 means that you're stuck with 8-bit 4:2:0 processing, then avoid it. However, I believe that you can take the original files into After Effects and process everything in a robust color space. If step 1 is to apply noise reduction and to provide 32-bit float 4:4:4 from there forward, you're golden.

One thing for sure, do not process in 8-bits. If you do that, you're guaranteed to see contour lines and other horrible artifacts, regardless of the codec.

Overall, Part 3 rocked. I would have liked to have seen more closeups of the keys from each of the cameras.

And, was it my imagination, or did the 5D2 have MORE DOF than the 7D on the text shot? The text from the 5D2 was sharper and had some color aliasing. The 7D was softer, so it aliased less.

The bottom line is that you can choose softness or aliasing with these cameras. A somewhat shallow DOF is the best way to avoid aliasing in the backgrounds. And in that case, the audience doesn't expect detail.