Just changed my debian box from squeeze to wheezy and set this up instead of poclbm. It's working well and much better than dealing with the hassles of the python-pyopencl packages dependcies on nvidia.

In my experience the miner makes little to no difference. But the settings you use definitely make a difference. Make sure you enable vectors - it makes about 50MH/s difference on my 5830.

Catalyst are the video card drivers and Stream is the OpenCL SDK (it's been renamed to AMD APP now though).

Cablesaurus is designing a chassis and planning to sell it. He also sells the extender cables you see in the photo.

I don't see the point either.
Running a second PSU means wasting more power - and that one only states above 75% efficiency. I personally won't use anything without 80+ certification these days.

Would also mean better airflow as they don't have to make room for the plugs and there's probably some things that could be removed from the board entirely, reducing the power usage and heat.

Not worth risking your career over IMO.

I understand the temptation though I do IT support using Kaseya (it's basically a botnet, but kaseya hates it when we say that) so could write a script to run it on 1000 machines in about 10 minutes... but I'd also be kissing my job goodbye - so in the end it's not worth it.

I'd just run them inside screen sessions, then you can connect from wherever and check on them.

Just something like:

screen -S "name of screen session here" -X screen  2>&1 1>/dev/null

One line like that for each instance of the miner and run it from whever, such as /etc/init/rc.local

I'm sure there should be a way to overclock from the commandline too, just have a search and run that command before spawning the miners.