These are two quotes I found on the web concerning BayerGreenSplit in Lightroom:

Eric Chan (LR4 developer): "You are right, the green split suppression is now being applied to all the Micro 4/3 models including G1 and GH1, so the definition is not quite as high as previously. We are still studying ways to improve the resolution and edge smoothness while balancing suppression of pattern noise."

Manuel Llorens García (rawness.es): "Adobe programs, ACR based (Photoshop & Lightroom) can effectively diagnose whether the image will produce artifacts or not and they fix it in the same way as dcraw, using something similar to a four channel VNG"

RawTherapee introducs pattern/labyrinth noise in some areas, too (see my last image). But not only do you have control over whether and how much Green Split/Equilibrilation you want to apply to overcome that, but the outcome is sometimes better to work with. The image before the last with the vertical fences shows considerable moire being left _after_ color noise filtering (default 25) has been applied in Lightroom. In fact you have to push the color noise filter considerble close to 100 to get rid of that while RawTherapee's filter has no problem with using low filtering.

And DXO still shows that the additional detail is possible without pattern and additional moire.

Thank for the hint, but my demonstration images 1,2 and 4 did not use any kind or sharpening or noise filtering at all.

Image 3 used noise filters and some unusual sharpening settings in order to pull fine line detail sharpness (to bring out the fences) without halos from all three programs. You normally wouldn't use these extreme settings to get "pleasing" results. Here are the settings:

DXO - Lens Softness

Global: -2.00 (to get rid of any white sharpening halos)
Details: 100 (for maximum contrast on fine line details)
Bokeh: 0 (to avoid blurring away finest details)

LR4

Amount: 37 (quite more than I usually use, but still low enough to avoid halos and reticulation noise)
Radius: 0.8 (to catch the finest details without having coarse lines blow over)
Detail: 100 (for maximum contrast on fine line details)

Masking: 3 (just enough to catch some noise in the sky background without blurring finest details)

RT4 - Deconvolution

Radius: 0.79 (too lazy to dial in 0.8 )

Amount: 75 (default, 100 would have increased some contrast on fine detail at the cost of noise, but I aimed to match DXO)
Damping: 0 (to avoid blurring away finest details)

Iterations: 30 (default, 100 would have increased some contrast on fine detail at the cost of noise, but I aimed to match DXO)

DXO's had some advantages over the other two contenders. First I deliberately chose a Color/Tonality profile that resulted in maximum contrast between the sky and the fence that resulted in the fence being slightly darker/more contrasty to begin with. And of course DXO's Lens Softness can be used to sharpen very strongly without increasing noise, because of the "hidden" and very effective noise masking done by DXO. I could have used more masking with LR4 and damping with RT4 to get a similar effect, but that would have blurred away some of the lesser dark/contrasty fence parts.

One thing I forgot to mention is that DXO's demosaicing is not entirely free of artifacts. There are vertical lines in several areas that are visible brought out by sharpening, but are already present in the demosaicing data. One example is the backside of the sign on the right side of image 3. The sign looks as if the vertical fence runs between the camera and the sign, but in reality the sign is placed closer to the camera.

That makes sense, especially since DXO seems to apply some form of masking or noise filtering during the demosaicing stage already.

I am least experienced with DXO, but haven't used any of these longer than when LR4 went public beta.  Which is why I even started my whole comparison to get an idea of how much difference it can make for certain things. Here is a thread where I documented my comparisons, with some full sized (non compressed) images and additional information about different tonality/curves and white-balance behavior of the different applications.

If you don't want to wade through a whole thread, here is just the images gallery (including the WB table): http://www.dpreview.com/galleries/3262748529/albums/raw-software-compa rison

Think I'm gonna start a dedicated thread about WB, too, as this one confused me.

Just saw your reply after posting mine. Exiftool is able to directly change the BayerGreenSplit field (-BayerGreenSplit=0) and then either write an output DNG to another file (-o filename) or simply overwrite the original one (overwrite_original_in_place).

To bad that this mostly results in creating labyrinth pattern instead of smeared away information. I would have prefered real information like both RT and DXO are able to deliver.