Okay, so, um . . . this seems fairly doable with existing hardware, only technically requiring firmware and/or PC link no?
As far as I could see you could do the easy ones by hand (stills):
Get a tripod, a fast lens with super fast focus motor . . . start with a reference shot with very high DOF. Then, set your aperture where you want it, and take repeated shots either using different AF points, or even manually focus . . . merge.
Granted, the reality of this is a slippery slope with regard to hand holding, automation etc; you know, genuine usability. Honestly, I think most of this could be 'faked' with a current Canon models, at least to the extent that you would ever need/want it for a real world situation.
On the other hand, instead of merging these and making them online, it'd make a nice sequential image in a frame
--EDIT--
I got to thinking about it even more and I think I get it, essentially they're talking about streaming sensor data 'realtime' (hah, no such thing, right?) and you can combine the fragments to 'fake' shutter speed, iso, no?
Another thought is if this is conceptually coming from a computer guy, I know where this is going . . . get as much data as possible and cherry pick the good ones . . . Maybe this is lobbying from the memory industry for bigger files