Hi Colin,
The Sony star eating algorithim is essentially the same as the one Nikon was using until a few years ago. It was actually fixed by a non Nikon hobbyist engineer Marianne Oleund.
Basically its an algorithim designed to remove hot pixels which are defined as too bright a pixel (compared to its neighbours within an 8 pixel section due to RGGB filters) that is within a 2x2 pixel grid.
It has not been improved really. A7, A7r, A7s have it where it kicks in in bulb mode (ie beyond 30 seconds) since about firmware 3.3 for A7r2 it kicks in earlier at any exposure longer than 3.2 seconds.
My A7r2 is running firmware 3.1 and should only be kicking in during bulb.
But the recent image I did with 5 minute ISO640 exposures did not show any evidence of star suppression - hard to tell when its a field full of stars.
From examples I have seen it mostly diminishes or erases dim stars in a field. So longer exposures may overcome it.
http://www.pbase.com/gregbradley/image/165750987/large.
The thing about Sony A7 is the high QE sensors and low read noise.
Sony A7r QE is 50% read noise of 2.3 electrons full well depth of 49714
Nikon D810 is QE 47% read noise of 1.3 electrons and full well depth 78083
I am not sure how Nikon coaxes such better full well and read noise out of the same sensor as the A7r.
A7r2 sensor data is not available but it would be substantially better than A7r.
Canon 6D has dark current suppression technology, so does 7D mark 2 which has high QE and low noise. The new 6D2 has much the same dynamic range if not very slightly worse as the older 6D.
My opinion is go for at least 45,000 full well. My experience with CCDs is that full well depth is an often overlooked stat of a sensor and one that affects the image quite a bit.
The Sony A7s stats are crazy good, 65% QE, 0.4 read noise and 155,000 well depth. If not for the star eater it would be perfect. But there are plenty of good images around from those using one so perhaps that should be the go. There is currently a lot of pressure being put on Sony to correct the star eater algorithim. So perhaps within a year it will be corrected but not yet. A recent firmware upgrade improved it marginally in the green channel.
A full list of camera's performance is here:
http://sensorgen.info/
Greg.