I like this idea, its very doable and would be very power efficient. I dont think you would want a basic Aluminium box shape though. Best approach may be to use a heat sink 200 x 200 x 20 minimum with a gap between fins centreline of about 10mm.
Trim each alternate fin by 15mm from its edge and do the same on the opposite side but on staggered fins from those first cut. Weld the end plates and then place a flat rubber mat (2mm thick) 195 x 195 on top of the fins and the closing plate on the top, then seal weld all around. Use a 10mm aluminium nipple for the inlet and outlet which will be on opposite diagonals. The water path would then "snake" backwards and forwards as it travelled laterally from inlet to outlet. The fins are important as they will transfer the heat evenly into the moving water like a conventional tube heat exchanger.

The outlet would be fitted with a rubber hose which transports the water well away from the rig  before it is squirted onto the top of a vertical (fine) corrugated plastic sheeting, as the water runs down, it is cooled rapidly by a few 20 W desk fans blowing cool air onto the corrugated surface. The cool water is collected in a simple plastic reservoir and pumped back up a rubber tube to the inlet via a submersible adjustable aquarium pump. PCB would have to be attached to sinks by double sided adhesive thermal pads and only use Aluminium nipples as any other different metal will cause galvanic corrosion and cause leaks. A low power small aquarium pump of say 20 LPM would do nicely. Flow can be adjusted until the heatsink outlet temperature is low enough.

It is something I might just try  

cheers,
kev

good job!

What about puting the hole thing in a fishtank and submerge in mineral oil to cool ??

a radiator of course to cool it back

I don't think that would really be a problem.  The PCI-e 6-pin connector has a design specification of 75W, and the pinout is:

Pin 1 - +12V
Pin 2 - N/C
Pin 3 - +12V
Pin 4 - GND
Pin 5 - Sense
Pin 6 - GND

So, looking at the wires actually delivering power you have 37.5W on each +12V GND pair at the rated PCI-e spec.  On top of that, the connector is actually rated for higher power delivery- more than double the 75W specification.  Assuming you use proper diameter wires and good termination, one pair of power wires will be more than enough.

tl;dr: yes, it would work ok.

(edit) Just noticed you had said pin 5- you would want to use pins 1 and 4, not pin 5.

You remember from avalon-ref "golden nonce=received nonce - 0xC0"?

thats exactly what i thought when i read his post.

I support this !

So the water is running through the inside of the "box"? The largest problem with any setup like this is making sure that there is conduction between the chips and aluminum to the water. There will be no convective cooling in this setup (such as blowing air over the chip without any heat-sink). I also wouldn't consider setting something like this up along with a radiator and fan as "easy" unless you already have this stuff around. But I would have to see a design to make sure I understand what you are talking about.

Edit: OK, I've seen a design similar to what you are talking about. So, these current designs are placing heatsinks/aluminum housing on the underside of the board... how is heat being conducted from the thermal pad on the chip, through the board, and to the heatsink/aluminum? If it is just silicon then there may be a conduction/heat-transfer issue. It seems that this QFN design isn't really made for external heat dissipation.

for efficient cooling with water better use small pipes to achieve a turbulent flow (see Reynolds number).
german users find all neede info in this book:
VDI Wärmeatlas

another aproach is to use heatpipes, perhaps also fanless.