Like for Forklift, the existing cap patch does nothing perceptible to my eyes (or my scope

). However, the 100nf cap going to Analog GND
greatly improves picture quality. That said, 100nf is not enough to eliminate some other artifacts like "ripple banding". See these comparison pics:
With a 100nf capacitor:

With a 10uf tantalum capacitor (as per the AD725 spec):

I've done some testing with non-tantalum caps and the results were pretty much identical (to my eye and scope again). So if you don't have some 10uf tantalum lying around, you can put any electrolytic cap from 4.7uf to 10uF (for both types, take care of the polarity!)
Here's my version of the fix with a 10uf tantalum cap. If you don't have tubing to protect the long pins from shorting your board, use some gauge 22 wire piece and just keep the shrouding (what I've done).

So, in conclusion, with this fix applied, I can confirm you have what is closest to the "Uzebox experience" (Forklift coined the term

). And I noticed no additional or significant noise on the digital lines. I realized my schematic mentioned the caps going to the digital ground, but in my wire-wrap prototype both AGND and DGND are the same. Hence that 'perfect' picture is what is expected. Note the best picture possible could be attained by implementing the luma trap on the AD725. That gives a sharper picture with less color bleeding when using the composite output. Perhaps there could be a way to hack that too on the Fuzebox for the ultimate experience!
Cheers,
Uze