DNG latest specs can be found here : http://wwwimages.adobe.com/www.adobe.com/content/dam/Adobe/en/products/photoshop/pdfs/dng_spec.pdf
Interesting link about DNG : http://www.barrypearson.co.uk/articles/dng/index.htm
Not so up to date (the author is not into cinema) info about cinema dng : http://www.barrypearson.co.uk/articles/dng/cinemadng.htm
Reading the specs (chapter 7), I find that we can add "opcodes" to the dng. They allow further processing to be described by the camera. Those processings take place after debayerisation.
For example lens distortion removal, vignette removal.
And more interesting for us, dead pixels removal / interpolation, to be done by the host dng reader. If I understood correctly, we define those deadpixels in the dng opcodes, and they are interpolated by the reader without additoinal intervention. As better interpolation algorithms emerge, we can reprocess the dngs to gain better image quality. Also, this way we don't need to do dead pixel removal in camera (except for preview and processed output), "just" need to mark them in the dng's.
Trimbounds is interesting as well, it allows to crop the image to a particular aspect ratio, while still recording the whole image in camera it would allow image crop according to the dop choice. Later cropping to another area could be performed.
DNG is mainly uncompressed. We could dig the compression they provide in the latest specs. I suspect it's some kind of zip compression, which would give us 1:1.5 - 1:2 compression ratio. Not so great.
Another option is to :
Note that compressing this way leads to non standard files, that need to be reprocessed back to dng on the postproduction workflow. Red code success is because they provide an end to end solution to this problem. I don't know if providing compression is worth it if no end to end solution exists for editing and post processing.