Indeed the xmega A1 series looks quite nice. And external RAM looks to me essential to fully take advantage of all the peripherals, DMA, event systems, etc. Speaking of the event system, I've looked at it but played with it yet. Though it seem it would work to stream out graphics in conjunction with DMA, I'm unsure about the DMA bus priority and the jerkiness if may cause if the main CPU can interrupt DMA transfers. We'll see. In any cases, I'm pretty sure we could do some pretty cool stuff with the event system and all of timers. Just the video timing could be entirely done in the background without any CPU intervention.
Also about overclocking, that's true that this was a lot of hassles for some. When DIP parts it wasn't too bad to switch it for another one when testing, but when using surface mount parts that's another story. Though my idea was to use a 3.579Mhz crystal and use the internal PLL clock multiplier to dynamically overclock the chip. So at least it would not be permanent. Btw, how stable are those PLL clock multipliers? Do they tend to drift making them unsuitable for the AD725?
As for the video output stage, I'm still thinking if all those xmega peripherals could allow me to generate the color modulation without the AD725. Since I suck big time at analog design, I may need some help at some point.
On the other hand if I stick with the 725, I'd like to change the DAC design to the approach you proposed a long time ago, using 2 bits for each component plus 2 bit for global "lightness". That would give better gradients and a much more usable color palette. Another option is to have a VGA output, but all older analog TVs would be left in the dark...-Uze
p.s.: What up with the Avcore and baseboard in 2011? Did we see the last batch at Sparkfun?
but I'm going to say it's in the 10-15% range (I built 128 boards; I think I have ~15 that didn't pass manufacturing test). For whatever reason, those QFN packaged parts always seemed more finicky than the DIPs. (ie, I wasn't even able to get video from the 1284P on the AVCore and I tried a few different parts)
I had already purchased Mega644's, PCB's and AD725's back when we first started though (which is good, 'cause 644's appear to be totally sold out at the moment), so I think I'll just see if I can run a few panels (just do like ~32 pcs) and hope for good yield. That'll be like a year's supply and I guess I'm only "out" the production costs since I already paid for the parts anyway. I should have 'em restocked by mid-February I think. Depends on if the assembly house is busy or not... They took 10 AVCores before Christmas and sold through those pretty quick, but it took 'em have a year to sell a few before that! Unpredictable sales to say the least. 
The idea of the PLL was to use a 3.579Mhz crystal and multiply it 9-10 times to extract all possible extra cycles and feed a 4x clock to the AD723 (since the AD725 is a 5V only part). I do some testing I guess. What kind of circuit did Andre Lamothe used for his XGS? From you comment it doesn't seem the quality was as good as on the Uzebox AVCore? Speaking of which, that's too bad you will restock them *after* their annual "free day" event! 
However I was wondering how do we do white/grayscale then? Since whites (in NTSC) is the absence of chrominance, shouldn't we need some sort of saturation bit or something? As for the overclocking, you're right, it caused enough issue so far, I'll keep it in spec this time.