Really? You think?!
Good grief: this is the sort of conversation I thought we'd nailed years ago . . . for anyone living in a cave since the 1900s, we should reiterate that the 16-35 Mark I is NOT a high resolution lens: demanding shooting on a Mark 1 1Ds revealed this a LONG time ago.
At the wide end, the 17-40mm is a little better, but nowhere near good enough for the 1Ds Mk2, let alone a 22MP version. Ditto the redesigned 16-35II. Of course the Mk3 is much more demanding; obviously poor results will be obtained using inferior lenses; for heaven's sake, yes, the lens is the weak link in the chain!
The 24-105L is a fraction better again, but we simply won't see what the Mk3 is capable of until someone, somewhere gets a half decent lens on the thing, at sensible apertures, with properly optimised RAW processing, and shows us what it looks like without JPEG compression. Until then, debates about IQ in extremis are futile – exactly as they were when premature babble surfaced in response to publication of Canon's Mark II samples.
Only a little more patience is required . . . .
Like all very high resolution full-frame cameras the IIIs is merciless in revealing lens flaws. This is particularly true with ultra wides and even more so with ultra-wide zooms. And since Canon ultra-wide zooms are among the (least good) of the breed, this is what that story is about.
I have the 16-35mm lens with me because I wanted to shoot some record shots of the seminar group at my gallery. These were done wide open in very low light. Naturally everyone wanted to look at the files, and they looked soft. Thus the issue. When Mark tried the camera with a much better lens the next day the issue went away.
And to answer the inevitable question, this was the original version of that lens and I have tried the new one and it's not that much better. Those who want to shoot very WA with high res Canon cameras and whose primary concern is image quality should shoot with primes.