[I]"What seems to be the most important for digital capture is to get the midtones exposed correctly. These seem to need to 'sit' in the right place on the sensor and data curve. If they don't, shifting them to the right place in post almost always seems to compromise the image more than (keeping shadow detail/protecting highlights) the benefits are worth."[/I]

If you're recording 8bpc and using linear gamma then yes. To state the obvious skin tones are the thing we're most sensitive to, I cannot speak for every human but I certainly pay more attention to human faces than the details in clouds however even for someone shooting ENG one can have your cake and eat it too. After all clouds that look like blobs in the background can be distracting.

The Basics Of Camera Technology from Sony explains how this is done in the current ENG cameras:

The mechanism of DCC is basically the same as knee correction (refer to “Knee Correction” ). The difference is that DCC allows a wider dynamic range by automatically controlling the knee point (and in some cameras, the knee slope) to a video level optimized for the scene that is being taken.

There's a more accessible copy of the document available here where one can download it as a PDF. Probably worth having on hand in case one day someone loans you a big camera and you're left wondering what adjusting "crispening" is going to do to the image. :)

Bob.

any credence to the theory that a softer, flatter, lower saturated pic is easier/better/cleaner for the h.264 compression to deal with? Yielding a "better" pic in the end, with less macroblocking or whatever?

Assuming CRF encoding in x264, there should be no difference.

When scanning b/w negatives, I get the full image range within 1-254, and scan in 16 bit mono at 1.0 Gamma. The result for importation into Photoshop is very flat.

Then set the input endpoints in Photoshop, often with "some" clipping in the upper whites, but not in the shadows. Gamma is raised to get some life back into the picture. Then curves to tweak.

Finally, the image is saved in 8-bit.

I apply the same principles to color video as much as possible, usually with consistent results. I've found that keeping all the upper highlights within range often flattens the output too much, without resorting to masking or burning.

For those that don't have a film scanner or a high end camera Kodak's Marcie is a good example to play with in Vegas.
I haven't had a lot of time to play around with her and V13 but so far I've found that yeah, SCS have finally fixed the Color Curves. Also the difference between what can be achieved using 32bit float's two different modes is quite dramatic, the "video" variant seems to clip highlights internally.

For those not familiar with Kodak's Cineon gamma curve in 10bit values 100% white is at 635 (~60%), between that and 1024 is all highlights that the nominal decode to 8 bit will clip. To recover them one ends up with a quite dramatic looking curve using Color Curves.

I can get her skin looking great but there's still some clipping in her hair, pushing the hair too hard and I get banding / posterization. Maybe the original 10bit scan couldn't extract everything from the camera original.

Bob.