Many of us are familiar with overclocking a CPU - driving its base clock speed and/or multiplier higher than the rated specifications. Overclocking is popular among folks who want to squeeze every bit of performance out of their rig. It has its risks, but it's generally safe to do if not taken too far.
Overclocking is possible because of the way CPUs are manufactured. Many CPU dies share a single silicon wafer during the manufacturing process. The exact number depends upon the size of the wafer and the size of the individual dies. After fabrication, the wafer is tested, as a whole, and the dies are given specifications corresponding to the lowest common denominator. One bad die lowers the ratings for its brethren, who may be capable of higher performance.
Marketing is another factor. Intel might get so good at cranking out 3.6 GHz-capable Pentium 4 wafers that they're not flagging any as 3.2 GHz. They still need to _sell_ 3.2 GHz chips, so that 3.2 just might be a 3.6.
If you try to overclock a 'bad' die, you might fry it or at least lock up your machine. On the other hand, if you overclock a die that's been underrated, you might have some headroom to play with. Don't get me wrong. Overclocking can cost you a CPU and sometimes even add-in cards which may be sensitive to CPU bus speed. Less is more, and you have to be willing to replace your hardware if it gets smoked. Plus, overclocking may shorten the lifetime of a CPU and/or cards.
So, we've covered how some CPUs are capable of more performance than the packaging might lead us to believe. These CPUs may also be driven at lower voltages with no loss of performance. Lower drive voltages translates directly into less power draw, longer battery life (for notebooks) and less radiated heat. The CPU is still running at its rated clock speed and multiplier, so there's no slowdown. The trick is to find out how low you can go with the voltage without crashing. There are two tools which help with this: RMClock and Prime95.
RMClock lets you mess with all kinds of CPU parameters (voltages, multipliers and the like) and Prime95 runs torture-tests. The trick is to NOT set RMClock to auto-start until Prime95 can run for a good half-hour without locking up your machine. Some say to notch down the voltage until you have problems and then kick it back up one step. Maybe that's good advice for a P4 Mobile heat box, but I'm kinda conservative with this stuff (and I have a Pentium M). I set my low-end voltage to 0.892V @ 6.0x (0.988V stock) and my high-end to 1.148V @ 18.0x (1.34V stock). I haven't yet defined any mid-range values.
I'm intentionally going light on the details because you want to read this before doing anything. It's important to understand what's going on, and the article is a very comprehensive introduction to the subject. Good luck!