I push exposures with PS ACR fairly often to recover highlights &
shadows, often doing two exposures & mask them together so the push on
one of those can be really extreme, enough to show the itty bitty losses
when it's amplified in PS like that, and printed to 13x19". Roger Clark
uses another obscure high-bit astronomer's software to convert RAW files
of tricky action wildlife shots into big poster size prints.

Iris is astronomer's software also and I've been tinkering with it,
trying to figure it out. I'll summarize what I've been able to make of
it, maybe JPS will clarify where I'm lost.

http://astrosurf.com/buil/us/iris/iris.htm

Click the camera icon to set your camera model before loading. Select
linear (I think this makes the image appear too dark, not sure). You can
set the WB on this dialogue box with RGB values but I have no idea how
to determine the right numbers!

File menu > Settings to set the 'Working Path' where files will be saved
out to & opened from.

File > Load Raw brings up the RAW file in grainy B&W bayer pattern grey
scale, zoom in to see the RGB pattern in grey.

The Threshold dialogue should be visible, set the top box to 4095 & the
bottom to 0. For astronomy work, you can figure out a dark frame
subtraction thing to set zero higher in some cases but I'll ignore that
for now. You can fiddle with these for the proper look of brighness &
contrast but since we're doing a comparison of raw values, don't worry
if it looks dark, that's the linear conversion. I have no clue what
color space this works in (ie sRGB) or how to control that.

Wave the cursor around & in the lower right status bar you see the raw
values from 0-4095 so you can see here if there is anything at all to be
recovered in highlights & shadows. I think you can split the image into
RG&B files to see which channel is blown or not. This is all before the
whole thing gets muddied up by merging the bayer pattern RGB.

From the Digital Photo menu select 'Convert a CFA image' to interpolate
the RGB bayer pattern into a normal color image. It probably looks too
green because there's twice the green pixels... I have no idea how to
correct this meaningfully, see WB comments above. From the Digital Photo
menu select 'RGB Balance' if you want to try... no undo that I can see
so it's frustrating.

From there just save as a tif & when opening in photoshop, sRGB seems
as good as any color space to assign. Then I do a levels adjustment to
set the brightness of the middle gamma level & make it look like a
normal exposure.

I probably got it all messed up but you won't find a more
straightforward explanation <g>. Most of the tutorial info on the web
site is geared toward processing multiple exposures of the same scene
for astronomy stuff and will make no sense to a regular photographer.

For your comparison photos, it seems important that they be comparable
exposures & I see a significant difference in brightness & contrast. A
brighter exposure will have less noise and more blown shadows. The
histogram on the camera is jpeg not raw but probably close enough to get
things real close. The two cameras histograms should have the hump
placed in the same spot & not be going off either side more than the
other to get comparable exposures. If that takes a +.3EC or -.3EC that's
worth noting but the exposures need to be the same.

I'm not criticising your test more than any other, it is *really
difficult* to do fairly (meaningfully) and I appreciate that you used
the same lens, I think zooming to match field of view (for the pixels)
works better now, just a slightly different range on the lens. I assume
the full image is larger on the D200 though so it might be fairer to use
the same zoom and enlarge the *MP version up to 10MP.

To compare what the cameras capture on a level playing field.  There are
so many variables when comparing conversions.  RAW data is *EXTREMELY*
easy to work with (mathematically, I mean - the tools aren't always
there); simpler than conversions by orders of magnitude, once you know
what it is you are looking at.  To use IRIS, you just take a shot with
two different cameras, set the same f-stop and shutter speed on both,
and scale the linear data so that the darker RAW image has the same
levels in its highlights as the brighter one, and you have a very simple
relationship between the two; the tones match.  You can apply a gamma to
both; the same gamma, and the tones are still the same.  If they're not,
then there is something we need to know about (high or low blackpoint
clipping).  In DPReview's D200 vs 30D comparison, the postage stamp
obviously has a completely different transfer curve applied with the
different cameras; the D200 postage stamp has much more contrast.

In one of my comparisons of the same camera at different ISOs, I had ISO
100 and 1600 both pushed to 10,000; I set the linear scaling of the two
to exactly a 16:1 ratio, the same as the inverse of the ratio of
exposure times, and the tones came out *exactly* the same; only the
noise varied.  I could never, ever do that with a conversion; not even a
16-bit TIFF.

I shoot with the expectation that there will be better software someday,
as RAW converters are still in the dark ages, IMO.

There is a place for comparing out-of-camera JPEGs, and a place for
comparing conversions, but there is also a place for what the camera
really captures - the RAW capture, and if you shoot RAW, you still have
this for future converters with better highlight and shadow handling.

If you compare based on the image output, you're actually comparing the
combination of camera and raw conversion.  Looking at the raw data
removes the conversion from consideration.

Since the raw conversion can't be the same from one camera to another,
especially if you use different software for each (but even if you
don't), comparing image output is much less meaningful.  If your
conversion for one camera happens to be inferior, you could arrive
at entirely the opposite conclusion from what you should; you might
say, "geez, this camera is way noisier than this other one," when in
reality, the interpolation and other software transformations may be
what is worse, and that camera might be much *less* noisy using other
software.