Wouldn't you be hard-pressed to get the data out of the ADC at those speeds, without an FPGA?
If the idea is that getting rid of the FPGA makes this a substantially easier project, I would suggest otherwise. At those frequencies, good layout, and keeping digital crud out of the front end, is the bigger challenge. That's what I heard from the openhpsdr folks (the best-known direct-sampling design).
Don't forget to factor into your supposed cost savings that you now need two ADC channels instead of one.
I also believe it's difficult to get that kind of bandwidth out of the Tayloe-style QSD, due to finite source impedance and reasonable capacitance at the four phase outputs, but I'm not an expert.
More interesting (to me) would be a moderate-bandwidth QSD-based receiver which is designed around fast and high-resolution successive-approximation ADCs, say 1 to 2 MSPS, rather than a sound card.