Lots of questions, of course, remain unresolved in my mind. First and foremost: How is exposure adjustment and highlight recovery handled and where is the best place to perfrom these adjustments? In DxO or in the subsequent DNG editor?
BTW. My first tests with regards to demosaicing comparing new DxO 5.3 vs LR2.1 show a very promising advantage for DxO (it was the other way around with DxO 5.2) with regards to detail rendering especially noticeable in details like grass, stony or sandy surface etc. Lots of blockiness and demosaicing 'meanders' noticeable in the LR rendition are missing from the DxO renderings which appear much more natural. Also small colour details in these surfaces are rendered convincingly by DxO while they have dissapeared in the LR rendering. That's obvious when viewed at 200-300%. Not sure if the differences would be noticeable in 'normal' use but the difference is certainly there. More tests to follow.
Above findings apply to Nikon D200 raws and not necessarily applicable to other cameras.
DxO 5.3 looks promising.
EDIT:
The differences looked too great to me, so I went back and redid my test paying more attention to compare apples to apples. It turns out that sharpening has a strange and unexpected effect to the results. Let me try to elaborate:
Case 1:
No adjustment at all made in DxO, DNG output, import in LR. Compare to NEF in LR with all detail modifiers to zero (sharpen, clarity).
The result favours DxO, although to a much slighter extent than noted above. It is quite tough to choose between the two renderings so for all intents and purposes I'll pronounce this a draw.
Case 2:
Take Case 1 and progressively apply sharpening in LR in equal amounts to both files. Here's the weird thing. The more sharpening is applied to the two files in LR, the more the LR rendering becomes preferable to DxO DNG rendering. The differences are not great but they are there with the DNG rendering appearing 'harsher'. Maybe LR sharpen algorithm is performed before de-mosaicing whenever the real raw file (not linear DNG) is available? That might help to explain the difference.
Case 3: Compare LR sharpening with DxO lens softness correction (for lenses where the DxO module is available).
Apply DxO lens softeness correction (Default at 0) before outputting the DNG. Compare in LR with NEF file sharpened with various combinations of the parameters in Sharpen adjustment. No additional sharpening to the DNG file. Consistently, the DxO DNG rendering looked better than the NEF rendering whatever I tried to do with the Sharpen adjustments. Now, that might tell something about my ability to use the Detail parameters optimally, but, still, DxO softness was at the default. For me, applying DxO softness correction would be preferable and less fuzzy as a method for initial 'capture' sharpening than LR Detail correction.
Conclusion
I will have no hesitation to use DxO as the demosaicing editor in case I need to use DxO corrections such as for geometric distortions and chromatic aberrations. In that case I will use DxO 'capture' sharpening rather than LR capture sharpening. I see no real reason though to use that elaborate workflow for all my images given the additional demands for storage the creation of DNG entails and also my findings in Case 2 (which will affect the output quality in case I use LR for output sharpening).
Next test will be noise handling.