http://www.youtube.com/watch?v=Csj7vMKy4EI
"Lookout Moon, America's gonna getcha -
gonna go kaboom - was nice to have met ya -
'cos ya don't mess around with God's America!"
Some time ago (nearly 10 years - wow!) I made a microcontroller-based homebrew MP3 player: http://codepuppies.com/~ben/sens/pic/mp3 . My big mess of wires was a tiny fraction of the size of his, and it caused me enough headaches - tracking down signal noise, random glitches, etc... Hats off to this guy.
I'd also recommend this book: http://en.wikipedia.org/wiki/The_Soul_of_a_New_Machine to anyone who finds the Wired article of interest. It doesn't get too technical, but it describes the trials and tribulations of bringing a new machine to life, initially on wire-wrap boards - as well as some of the politics involved in a corporate environment.
The difference between 8-bit and 12-bit when working with DSLRs is actually pretty significant. You can pull a surprising amount of detail out of underexposed regions with 4096 shades that would otherwise be lost with 256.
Rounding errors (generally banding artifacts) will also be much quicker to appear with lower bit depths. If you're going to support 16 bits per channel, one may as well go all the way and support higher bit-depths too.
"I think trash is the most important manifestation of culture we have in my lifetime." - Johnny Legend