Hi,
i have not seen such a thing before, so i will pass this information on to the design team to see if they have any insight into this. But looking at your data, it looks like the when you get the errors it is always a matter of getting a code that is 256 larger than would be expected in that spot. And not just getting a '1' in the 9th bit position where there would have been a '0', but sometimes where there is a '1' in that position getting a '1' added to that position which ripples uo into higher bits. hmm. i don't see what that would be.
What are you using to catch the sampel data from the ADC and deserialize it?
Regards,
Richard P.