Each of the 3 power rails (3.3V, 5V, +12V) are connected to a fuse, then to a binding post, while the negative binding posts go to ground.
This sounds like the problem. Keep in mind, voltages are differences in potential between two points. There's nothing special about ground, it's just an arbitrary point which we pick. It's 0V because the difference of something with itself is 0.
You can call anything you want "ground". That's why this circuit works. If you call the -12V output as "ground", then everything else is 12V higher. That's including what was previously called "ground": now it's 12V, because it's 12V more than what you are now calling ground.
Now consider what you've done:
simulate this circuit – Schematic created using CircuitLab
The "ground" of the power supply connects all of the voltages supplied together (connections labeled B and C). The output voltages are relative to this. Notice how the -12V (V4) makes a negative voltage because it's positive side is attached to "ground".
Then, you attached the negative binding posts of all the supplies together. Largely this is redundant: you are duplicating connections B and C. But you are also adding connection A.
See the problem? You've shorted out V4. A wire has ideally zero resistance. By Ohm's law, the current that will flow is:
$$ \frac{12V}{0\Omega} = $$
In reality, the wires used to make this connection actually have some very small resistance, and you get a whole ton of current. This far exceeded the current the voltage regulator can handle and the smoke got out.
Another approach is to say to Hell with elegance and just brute-force it. Get a bunch of power resistors and hook them up. For instance, DigiKey will sell you 5-ohm, 10-watt resistors for about $1.60 each. Each resistor will draw 1 amp at 5 volts, and 30 of them will cost about $50. Set up an array with some way to connect them independently and you can provide your load in steps of 1 amp, which I'd guess is fine enough resolution for your purposes. It's crude, but probably quicker than messing around with transistors and heatsinks and all that.
Best Answer
As I understand it you shouldn't need one. ATX supplies are supposed to be well regulated.
See
Some Intel ATX specifications say
As noted many many websites do say that a small load is required. If you want to follow that advice, use a small power resistor. A desktop PC probably uses 2-3 W in standby so something approaching that should probably be your target.