The duplicator's ADC is 10-bit unsigned; the FPGA converts this to scaled 16-bit signed data (as 16-bit is the only thing the USB3 can pass - you have to transfer 16-bit anyway). Since there is no processing of the sample by the Linux app, that is what is stored on the hard drive.
So you have NTSC capture at 28.63 MSPS @ 16-bit and PAL capture at 35.47 MSPS. The size of the sample depends on the length of the disc, so it's approximately:
1 second NTSC = (2 bytes * 28,630,000 samples) = 54.6 MBytes 1 second PAL = (2 bytes * 35,470,000 samples) = 70.9 MBytes
So if you sample 60 minutes of an NTSC disc (one side of a CLV) you end up with about 200 Gbytes of data or 250 Gbytes for PAL.
Since (as someone pointed out earlier) the disc is pits and troughs the most important thing you get from ADC resolution is actually sensitivity across a wider peak-to-peak RF range (as the signal is weak at the beginning of a disc and strong at the end). So 10-bits was the compromise between need and cost (ADCs get a *lot* more expensive above 10-bits).