SOLVED 
I was having this same problem and it was driving me nuts.
I had tried installing 1.5.2 on three different machines (XP, 2x Win7), re-installing 1.0.5 and loading code to a 2560, rebooting multiple times, replacing the 16U2 firmware (2 other versions), different USB cables - but nothing worked (except manually erasing and resetting the board).
The other clue was that even when the DUE was recognized and programming proceeded - loading a sketch failed to reset the DUE (manual reset was required a second time).
THEN, I read the post from Aiekakiku Errors in Due board/schematic files - Arduino Due - Arduino Forum and decided to try it.
I shorted out R23 with a small flat blade screwdriver (pressed on the side of R23).
It worked!
I then soldered a short across R23 - and it has continued to work on both my Win7 and XP machines, every time.
The short is not the safest solution and I will try a 1K resistor as a permanent fix.
So this appears to be a design flaw in the DUE. R23 is the wrong value.
Specifically, natural variation of R23 and the 15K pull-up on the SAM3X can result in inadequate margin to the input threshold (Vil) of the NRSTB signal on the SAM3X.
This means some systems work OK, some are intermittent and others fail all the time.
I'd be interested if this fixes other DUE boards that have the same problem.
Many thanks to Aiekakiku for his original post.
Andrew