The difference in DR between the A7R and D810 is due to a new iteration of the sensor design which has a lower minimum hardware ISO setting, nothing to do with RAW compression.
ARW compression first maps the linear 14-bit image into a non-linear 11-bit space. This does not compromise dynamic range as more bits are allocated to the shadows than the highlights (due to the eye being very sensitive to similar shades in the shadows and almost blind to similar highlight shades).
The second part of the compression takes 16 pixel subrows, identifies and stores the smallest and largest values, and then stores the other 14 values according to where they lie between the largest and smallest, using 7 bits each (128 'levels').
When you have a edge with a very large brightness difference 128 levels is not sufficient to store subtle variations in the in between values, hence you get some degree of posterisation (several values mapped to the same number), which can affect DR, but only locally, within the 16 pixels.