I too used to look for reasons to believe that DxOMark was doing honest, but, flawed testing. The more I delved into their site and their data the more I came to believe they are just Nikon Marketing Shills.
As for your comments on the superior noise capabilities of Sony sensors and the D800. The 5D MkIII has more accurate ISO, better SNR 18% scores (screen), better Tonal Range (screen), and equal Color Sensitivity (screen). All of these are components of Dynamic Range, and the D800s Dynamic Range is a straight line just like all the other measurements. Yet, the 5D Mk IIIs Dynamic Range gently levels at around 11 EV (screen), just like every other Canon camera using the CR2 file format.
It is my contention that it is NOT that Canon has not improved the Dynamic Range performance since they implemented the CR2 file format, but, that DxOMark is unable to decode the portion of the CR2 file that contains that additional Dynamic Range information. It is further my contention that the ‘pattern noise / banding’ that began when Canon dSLRs started using the CR2 file format, is actually where that additional data is stored / encrypted.
First off, I do not believe that banding and FPN is restricted to the CR2 format...look at some CRW's and you'll see similar problems in older sensors, just to a lesser degree (which is not surprising, given that older cameras had much larger pixel pitches and QE that is only marginally worse than what Canon offers today.) FPN and banding is not even relegated to Canon...its a problem inherent in most digital sensor designs except Exmor!
The reason Canon levels off at low ISO is because their read noise jumps! At ISO 800+, Canon read noise is about the same as Nikons...3 to 4 electrons worth or so per pixel. At ISO 100, Nikon's, thanks to its Sony Exmor sensors, is STILL around 3 to 4 electrons...but Canon's spikes. At ISO 400, Canon's read ISO on the 5D III is @10 e-, at ISO 200 it is @18 e-, and at ISO 100 it is @33 e-. Even if you factor in the bias offset properly, with 33 electrons worth of read noise, your bias adjustment is going to be very small or zero (so that extra 2048 levels worth that Canon reserves as a bias offset is REQUIRED to raise the black point above the noise floor...and even then, its obviously not enough as electronic noise, including pattern noise forms, still exhibit at ISO 100, 200, and 400.) The reason Canon sensors develop a "shoulder" at the low ISO end of the DR scale is because of that read noise curve.
I'm not a Nikon fanboy, I don't care much for their ergonomics nor their menu system. I do not own any Nikon gear, my entire kit is Canon. I don't care much for DXO, to a certain degree they definitely do pander to Nikon, which isn't surprising as Nikon is one of their major sources of funding. The DR curves, though, are not fake or some scam perpetrated by DXO (although I do believe their Print DR results and anything based on them IS a scam). There are a variety of other, independent sources that produce roughly the same DR curves for Canon sensors as well. Read noise is not a joke, and it is, most definitely, a DR killer at low ISO.
Once you start using higher ISO settings, the physical nature of light kicks in and produces noise on a level far beyond what the electronics themselves can produce, which is why pretty much all cameras, regardless of brand, perform about the same at ISO 800+. Its not that Canon has done anything particularly special there...they don't have to
to be competitive at high ISO.
Seriously!!! do you honestly believe that almost every single Canon digital camera using the CR2 file format has almost the exact same maximum Dynamic Range (as tested by DxOMark - screen).
Every single Canon camera that DxOMark has tested except the PowerShot G9 and G10 have the same maximum Dynamic Range (screen) of around 11 EV. Canon can produce a 1/1.7” sensor with 11 EV (PowerShot S90 and G11 and above), but, they can’t produce a full frame sensor with more Dynamic Range than that?
That the Canon Digital Rebel (D300) and the Canon 5D Mk III have the same maximum Dynamic Range?
That it’s all related to sensor tech and has nothing to do with the CR2 file format?
So, Canon was able to increase the Dynamic Range of the 1/1.7” sensor used in the G9 and G10, but, they couldn’t increase the Dynamic Range of their APS-C or full frame sensors since 2003?
Give me a break!!
Seriously!!!, all these Canon cameras have basically the same maximum DxOMark Dynamic Range 'measurement' (screen 10.9 EV +/- .1 EV < that's point one).
1D Mk II
1D Mk IIN
1D Mk III
1D Mk IV
Yeah, it has everything to do with Canon's sensor tech and nothing to do with the CR2 file format. NOT!
So long as Canon does not address their low ISO read noise problems, every single Canon camera that uses sensors manufactured by Canon will have roughly the same low ISO DR problem as well. The problem scales across sensor sizes as well, as the smaller your pixel well is (barring any additional mitigating factors like a backlit design) the lower your SNR, and therefor even if you have a lower read noise, on a relative scale its about the same for every Canon sensor produced in the last 4-5 years. The 7D has lower read noise than the 5D II or III, however it has about the same DR curve. Its a denser sensor though, with a considerably lower pixel saturation point. The ratio of read noise between the 7D and 5D III is 8.6/33.1, or around 0.26. The ratio of pixel saturation point between the 7D and 5D III is 0.29. The read noise to saturation ratio for the 7D is 20178/8.6 or 2346, and for the 5D III is 67531/33.1, or 2040 (which partly explains why it edges out the 5D III DR at ISO 100). Similar ratios all around. Canon designs their sensors in a certain way, and they have not been particularly innovative over the last four years or so when it comes to their sensor design (perhaps too confident in their own superiority?? Now caught off guard by the innovation of their competitors??) The fact that the relative results between these two cameras are roughly the same despite being generations apart and different models is not surprising!
And, frankly, Canon needs a kick in the ass to get their own engine of innovation back into gear so they can claw themselves into the modern era of digital camera design.