The 1D X definitely has better noise characteristics compared to the others, and it does not seem to exhibit the oscillating +/- 1/3rd stop ISO quirk (which is a big improvement IMO).
That is nothing new. All the 1 series, certainly any half-way recent one has used two amps, 1 for the main ISO and then one to adjust 1/3 stops on top of that. 1D4, 1Ds3, 1D3 were all free of that too. No non-1 series has ever been free of it though, they all used faked 1/3 ISOs.
(I am not sure why it works that way, but whatever the core, common, fundamental design factors of Canon's 500nm process are, it causes a non-linear read noise issue in all of their sensors (basically, read noise climbs from a floor around two electrons worth at high ISO to between 10 to 40 electrons worth at low ISO. Sony Exmor sensors effectively have a flat, linear noise curve. I'm hoping a move to 180nm process will allow Canon to solve those issues.)
A combo of the process size (since D4 sensor doesn't use exmor but does have less read noise than any Canon) and lacking the column ADC, more fully digital on chip 'NR' of the exmor and other sensors of that sort of design.
(some speculate those types of designs can't be done on 500nm process though, even with patents, since they take up too much room at 500nm, some speculate that the patents are not even an issue at other and that others do similar things and that it is the size alone that might be holding Canon back, not sure though).