The in-camera JPG comparison is a valid test if you shoot JPG. I don't, I use Lightroom and RAW so I downloaded the samples from Imaging Resources and DP review and processed the RAW files. I then processed all the files identically and up sampled them to the same size as the Pentax 645D. I then printed some of the files at 13X19" to see what actually mattered. (lots of ink and paper so I didn't print all ISOs)
5D Mark II and 5D Mark III are virtually identical.
D800 is slightly better than the 5D.
P645D is better than the D800. The difference with the D800 is more noticeable than the difference between the D800 and the 5D.
The above results are visible both on the screen at 100% and visible on prints at 100 ISO. The sharpness results wouldn't matter if you had even minor focus error or a narrow depth of field.
Hard to see in a well processed image or on a print, but the Pentax 645D first, then D800 then 5D Mark III at 100 ISO. At higher ISO very hard to see the differences although noise is related to dynamic range especially at higher ISO.
5D Mark II has about 1 stop more noise than the 5D Mark III
D800 has between 1 stop and 1/2 stop more noise than the 5D Mark II
High ISO 13X19" print limit
5D Mark II ISO 6400 and ISO 12,600 in a pinch.
5D Mark II ISO 12,600 and ISO 25,200 in a pinch.
D800 ISO 6400, and ISO 12,600 isn't too bad but ISO 25,200 is pretty ugly at 13X19"
Your tolerance for noise may vary, but the relative position will probably be the same. It really doesn't matter to compare the noise up sampling or downsampling since Lightroom does a really good job of colour noise reduction. Rankings stay the same.
None of these results are surprising to me as sensor efficiency is so high almost all differences between cameras can be explained by pixel, sensor size and photon noise. It is disappointing in a way because it also means there is little room for improvement is raw pixel sensitivity left. Most advances are likely to some from processing. Another good reason to shoot RAW.
My opinion, 36 mpx probably isn't worth the extra file size and effort but it doesn't hurt all that much either. For the landscape guy the extra pixels may show up but they would be better off with medium format for more resolution. There are other improvements in IQ with Medium Format due to less enlargement needed. For Nikon users I have noticed the D4 and the D800 are a significant improvement in resolution. Canon users have had that resolution since the 1Ds Mark III.
Lower noise probably isn't worth the 5D Mark II upgrade but my opinion is that the 5D Mark III is worth the upgrade for the improved focus, better body and other stuff which Canon probably could have improved with a firmware upgrade, like better ISO features. The 5D Mark III now has all the features I really like about the Pentax K-5, except the smaller body size. The 5D Mark III is a great camera to use.
Bottom line. The 5D Mark III and the D800 have great IQ and the 5D Mark II isn't too bad. The D800 resolution can be noticed, but it isn't the same as Medium Format.