A few things you can be sure of:
- Face recognition (with the intent to assist focus) requires live preview. DIGIC III needs a live image to determine where the faces are and adjust focus accordingly.
- Real improvement of dynamic range is achieved by improving the image sensor itself. Each pixel needs to have a bigger difference between Empty and Full.
- Increasing the bit-depth of the output signal, and not just empty upscaling of the digital signal, requires better ADCs (Analogue to Digital Converters). These are also embedded on the image sensor itself, so again this would come down to improving the image sensor itself. This has nothing to do with the processing chip.
- DIGIC, as a concept, has always been a multi-purpose chip. It's built from the ground up to work with different sensors. The orignial DIGIC sat in dSLRs and in digicompacts alike, G3 and 10D for instance. DIGIC III sits in the Powershot G7, expect it in new dSLRs.
- DIGIC III has support for the newer DDR SDRAM. Use of DDR instead of SDR SDRAM will double the I/O speed to buffer. Oh, and DDR is cheaper than SDR nowadays. Another reason to put DIGIC III in new models.
As far as powerdraw and other benefits: You can expect (but that's not a guarantee) that Canon has been able to move to a smaller node in the lithography process. It would be very stupid to stay at a big production process. Smaller chips cost less to produce.
Just look at the processor industry, AMD and Intel: Moving from 350 nm to 250 nm to 180, 130, 90, 65 and 45 nm has increased performance per watt tremendously.