I would definitely want an external USB DAC rather than an internal PCI-E soundcard.
I think you miss the point a bit. We already have very high quality DACs in the Casablanca and, in Steve's case, the Gen 8 III. At the moment, and I don't expect this to remain so for too long, we need to feed the Theta products S/PDIF. For now, Theta does not have a USB input card. So, one either heads down the path of a high quality sound card that outputs high quality S/PDIF from the audio server or deploys a USB->S/PDIF converter such as the Alpha USB. Personally I'm not ready to jump to the expense of the latter just yet because I fully expect Theta to provide a USB input card in the not too distant future and an integrated solution (isolated USB input with i2S to the DAC) should
provide a better solution than the Berkeley product. I do agree with you, as I've said above, that the deployment of a separate USB card in conjunction with the Alpha USB is likely redundant. BTW the Juli@ XTe card is galvanically isolated by design so no ground loop issues there.
Sample rate support over HDMI isn't a processor issue (as I understand it). Rather, at least in the case of the motherboards I've explored it depends on the RealTek chip and codec deployed. The Intel DH61AG uses RealTek ALC892 as does the motherboard recommended in the "Zuma" spec. The ASRock Z87E-ITX uses RealTek 1150. You can view the specs of these on the RealTek website. While they are marketed as "up to 192kHz" capable, a closer read of their specs shows they support 44.1, 48, 88.2, 96 and 192 but not, strangely, 176.4kHz. Oddly, however, if you look at the data sheet for ALC898
there are plenty of references to 176.4 capability even though the marketing specs do not include 176.4 in the list of supported sample rates. Personally, I've not yet had the time to directly test 176.4kHz on my setup.
My research suggested that USB out, using the SOTM card and a quality USB to digital converter (BADA is Da Best) will sound better than using the Juli@XTe. And the Bryston BDP-1 I had used for two years has a Bryston modded Juli@ card.
That's good - it's what one would expect! A properly isolated USB connection with the data then properly reclocked distances the Alpha (and Theta DACs) from the source device - the Alpha is the source. A direct and properly implemented USB feed into the Casablanca would be even better. An i2S feed from the audio server to the Casablanca would be even better as it would avoid the conversion to USB. But I will bet the SOTM is redundant. Remember also that the Bryston BDP is about the most basic implementation of an audio server possible. They use the older PCI version of the Juli@. It is driven by a very basic power supply - the motherboard does onboard the conversion to 5V and 3.3V - and they've made no attempt to improve on the Juli@ card (power it independently, upgrade the clocks and capacitors etc). The only "mod" they've done is to avoid the breakout cable with a jumper off the Hanrun part which provides isolation and add a 75->110 Ohm conversion board to support AES/EBU. Anyone with a soldering iron can jumper off the Hanrun to 75Ohm BNC or RCA and avoid the breakout cable. Beating the Bryston (except in shear convenience) is very easy.