Saving the sky and clouds usually doesn't require underexposing by 1.5 stops. Are you using Highlight Tone Priority? That saves some of the sky/cloud highlights. Are you using any fill? Use just a little bit of fill flash (or a reflector) and you are reducing the need to brighten a photo in post. One stop of brightening is usually perfectly OK, while 4.5 stops of brightening is not OK. Requiring a photo to withstand 4.5 stops of brightening in post looks to be the result of a combination of mistakes.
Finally, I wonder about the processing technique here. Was any noise reduction applied?
Really? I typically run into skies where I have to underexpose by 3 stops or more.
And 4.5 stops is not OK? But it is on many cameras today; that's my point. Let me put it this way: do you ever shoot at ISO 2200? B/c that's ISO 100 underexposed by 4.5 stops. So if shooting around ISO 2000 is acceptable to you, then underexposing ISO 100 by 4.5 stops shouldn't be 'not OK'. You may as well as criticize everyone who shoots above ISO 1600.
Also, I underexposed by about 1.5 stops compared to what the meter would've chosen here, IIRC. The extra 3 stops came from vignetting - so I didn't even choose to underexpose by 4.5 stops here...
Following this discussion, there are good points on both sides.
I'll take every bit of improvement a camera can offer, including 18 stops of DR if they can do it, so that's great that the D810 sensor gives a bit more leeway. But that doesn't define the entire system, both have their advantages. The Canon 5DIII is an amazing all-round performer that does better video, without moire or aliasing, better low light performance, excellent color rendition, excellent AF, a better selection of high end lenses (IMHO), better ergonomics (again IMHO), and that's before we talk about magic lantern or the fact that the 1Dx is deadly or that the 5DIV will be here long before the next Nikon.
Either system can deliver great results, and in most cases indistinguishable. This sometimes sounds like a left-twix vs right-twix war.