I would have sworn that my EOS M was producing better RAW high ISO shots then my 7D, yet comparisons at DPReview and Imaging Resource do not confirm this. So it looks like there really isn't much change among the 18 MP sensors from Canon. (My initial impression was probably due to the exposure being consistently optimal on the M. I shoot manual on the M and it is incredibly fast and simple to ETTR off the screen.)
The 70D is Canon's first truly new APS-C sensor in a while, and to me it does look a bit better in RAW. But not by much, maybe 2/3 stop?
On the flip side, for all the wild claims about Sony and Nikon sensors, they look no better. And they also show no difference as you flip through recent camera model iterations. Yet we constantly hear how Canon is "stuck" with old sensors, and how much better Sony/Nikon sensors are at high ISO. The comparison images expose the myth. A myth which is driven entirely by DxO nonsense.
I think the whole industry is at a point of diminishing returns on high ISO when it comes to crop sensors, though a new fabrication technique or other technology could change that in the future.
To be honest...and this is going to ruffle feathers...there isn't that much difference between the crop sensors and the 5D2 either. It's there, but not huge. The IR samples show it better then the DPReview samples (not sure why). From the times I've shot with a 5D2, the difference is amplified if exposure is less then optimal. And 5D2 RAW files are able to take more NR at high ISO.
The 5D3 and 6D improve on the 5D2, and to me the jump from crop to 5D3/6D is much more substantial in terms of high ISO. For that reason I tell crop shooters who want better high ISO to skip used 5D2's and go straight for the newer models.