TK,
Keeping the heater leads pressed against the chassis keeps the radiated field at a minimum, since the big ground plane is there to suck it up. This is generally the best approach. Another method which can work in some cases is to fly the leads high above the sockets, keeping them far away from the rest of the wiring. This distance reduces coupling into the signal circuits. That's what was done in Fender guitar amps. In either case, the main thing is to keep the heater leads away from the signal wiring.
Electromagnetic and capacitive coupling is greatest between parallel conductors and least between perpendicular conductors, which is why (as PRR points out) when signal leads must cross the path of heater leads, they should do so at right angles.
As for how a "humdinger" works, let's start by exploring why a heater circuit should be referenced to circuit ground in the first place. When the heater supply is floating, and it couples into the signal circuits (usually via leakage to cathodes, since that's the nearest electrode to the heater), its "return" to the low side of the circuit is strictly through electromagnetic and capacitive leakage paths. Without an electrical connection to fix the amplitude of the heater voltage to 6.3VRMS from the amplifier's perspective, a large amount of hum is induced.
Anyone who's touched a finger to the tip of a cord plugged into a guitar amp, or forgot to hook up the ground clip on his scope when looking at a circuit that didn't already share the scope's ground reference, is familiar with the amount of ruckus that occurs with a floating input voltage.
Grounding one side of the heater will reduce the noise considerably, and indeed that's as far as they chose to go in some designs. But why not take the next step and try to cancel it out even more? If, instead of connecting your reference to one side of the heater, you instead place it halfway between the two sides, you will make the hum voltage differential with respect to the cathode (or grid, or whatever else is picking up hum from the heaters)--in other words, two signals of equal instantaneous amplitude, but opposite polarity. So they cancel out. Some choose to do this with fixed resistors, others use a pot to allow you to fine-tune the balance.
The heater current does not flow through this reference--it just flows from one side, though the heater, and returns to the other side. So the actual amplitude of the heater voltage across the heaters is not affected, but its amplitude as seem by the amplifier can be reduced nearly to zero. Essentially, since the heater voltage is always going to couple into the amplifier to some extent, much better to inject two half-amplitude but opposed-polarity hum voltages rather than one full-amplitude hum voltage.
The resistors (or pot) do sit across the heater circuit and therefore draw some current, however, which is why the value of resistance must not be too low.
Fixing the reference a certain number of volts above ground does have value in preventing heater-cathode breakdown in circuits where cathodes are sitting at many volts above ground. Most tubes cannot tolerate, for long, a large differential between the heater and cathode voltages. But it's no good to "bias" the supply as a knee-jerk reaction; it really gives no benefit in circuits where the cathodes are only a couple of volts above ground, and could actually do harm by pushing conditions closer (in the other direction) to heater-cathode breakdown. If you look at tube datasheets, you'll often see both a positive and a negative rating for maximum heater-cathode voltage.
[Just before hitting "Submit"]: Man, I could probably edit all that verbiage down to two paragraphs, but it's late, I'm tired, and I'd better hit the sack...
