AVR Freaks

It depends how many meters your voltage regulator is from AVR. If they are within 1 cm, maybe you don't need it. It does not hurt to put it there. Who knows without measuring. I would put it there.

Not noise, not from PCB tracks itself. But PCB tracks are not 0 ohm superconductors, they have resistance and inductance. Now AVR or any other digital chip connected to same PCB tracks consume current in peaks, and current peaks turn into voltage peaks in inductance and resistance.

Can you get away without decoupling caps? The quick answer is 'maybe.' The problem has to do with the way digital circuits are designed. The typical 'totem pole' output has a pull up transistor to make the output go high, and a pull down transistor to make the output go low. At any given time, one of those transistors is on, and the other is off. The problem comes when the output changes state. Transistors tent to turn on faster than they turn off, so there's a brief period when both transistors are on. This creates a short circuit across the power supply for a few nanoseconds. With microprocessors, there are hundreds of logic gates, all synchronized by the clock to change at the same time.
The problems come when these very large, but short current spikes are being drawn through the resistance and inductance of the power supply wiring and circuit board traces. This causes the power supply voltage at the microprocessor pin to drop, potentially enough to reset the processor, or cause random data errors. If the power supply regulator is close to the microprocessor, you might get away without the decoupling capacitors. If it's not, you'll get random intermittent resets and crashes that are almost impossible to troubleshoot.
The bottom line is that unless you've got a very, very good reason not to put the capacitors in, saving a few pennies in capacitors just isn't worth the risk of having to deal with the problems lack of decoupling can cause. My rule of thumb is a 10uF or larger electrolytic capacitor at the regulator, or the point where the power enters the board if it's an off board regulator, and then a 100nF capacitor for each chip, located as physically close to the chip as possible.

That makes sense. I don't often deal with chips having multiple supply pins

gr8 is right. If you have other chips using the same vcc, especially analog stuff, the decoupling caps are there primarily to keep the noise at the uC, being able to provide voltage support for the uC is a nice side benefit.

To increase the effectiveness of decoupling caps at isolating VCC noise, have thick and short tracks from the uC VCC pins to the cap, then run long and thin tracks to the power supply VCC. The uC will get most of the transient power from the cap due to the low resistance path to the cap. I am not a big fan of having VCC flood planes on PCBs, you have no control over the impedance between the decoupling cap and the power supply VCC.