All has to do with relative specific heats, i.e., the heat capacity per unit volume at a givern temperature. Water has a far higher Cp than Air, in fact air is more of a thermal insulator. Most wall and roof insulation only acts to create pockets of "dead", i.e., non-circulating air, 

Now if we could only find a fluid with the heat capacity of water, but non-conductive...

Bet they're pussies when it comes to overclocking...

Supercomputers have been watercooled since long before it was fashionable.

I dont see what the big deal here is. Other than it being one beastly computing unit.

I believe the last milk-cooled super-beast was booted in Curdistained. Oops! I've gone and spilt it now! [sob]
Sod.
IBM has been using chilled-water since they were building those huge 3270 locomotives, and inventing DP, and other such dervish.

So what happens to the computer when the micro pipes start to rust and leak? The video says that the water is as close as possible to the CPU cores. Cool idea, but it seems like it could short an entire rack at a time.

The video has a clever thing in it: using the water cooling waste heat for heating in people's radiators, hot showers etc.

This is good because a standard heater would use a lot of power without actually doing any computation, in principle you could get the computing almost for nothing provided you wanted a certain amount of heating.

Water cooling is a bandaid. Why isn't IBM fixing the power consumption/ heat challenges at the CPU level? A rack of p 575's which is 448 Power6-cores @ 4.7 GHz consumes 71.7 kW !! At the rate that IBM's Power CPU developemnts continue to consume power and generate massive heat, they'll need more than water to cool these hot chips down!