Hi,
The gaps are just BS. Let's assume that we have saturation at DN (Data Number 16000). Shot noise will have a standard of deviation of 126 DN. So the value will vary between 15872 and 16128 (if I recall it right 65% of the values will fit into this interval). If we would bin eight of those 256 values in a single bin it would represent the distribution exactly. Exactly in the sense of the quantisation error being an order of magnitude or two less than statistical variation. Using a lot of bits for large values is just meaningless. There is a proverb for it: 'Measure with micrometer, mark witch chalk and cut with ace.'
The Sony cRAW format uses another form of compression. It uses 16 or 32 pixel samples, saves minimum and maximum values and uses 7 bit coding within that interval. That coding may have problems reproducing very steep gradients.
Best regards
Erik
I'm still none the wiser 
Is it raw lossy compression that causes grain/orange peel or is that just caused by un optimised processing?
Are there really less levels of information effectively making the Raws only capable of 8bit images and having steps/gaps when you process them in 16bit?
