From Paul White's "Take It To The Limit" on SoundOnSound (
http://www.soundonsound.com/sos/1996_articles/apr96/compression.html)
Peak/RMS operation: Every compressor uses a circuit known as a side chain, and the side chain's job in life is to measure how big the signal is, so that it knows when it needs compressing. This information is then used to control the gain circuit, which may be based around a Voltage-controlled Amplifier (VCA), a Field Effect Transistor (FET) or even a valve. The compressor will behave differently, depending on whether the side chain responds to average signal levels or to absolute signal peaks.
An RMS level detector works rather like the human ear, which pays less attention to short-duration, loud sounds than to longer sounds of the same level. Though RMS offers the closest approximation to the way in which our ears respond to sound, many American engineers prefer to work with Peak, possibly because it provides a greater degree of control. And though RMS provides a very natural-sounding dynamic control, short signal peaks will get through unnoticed, even if a fast attack time is set, which means the engineer has less control over the absolute peak signal levels. This can be a problem when making digital recordings, as clipping is to be avoided at all costs. The difference between Peak and RMS sensing tends to show up most on music that contains percussive sounds, where the Peak type of compressor will more accurately track the peak levels of the individual drum beats.
Another way to look at it is to say that the greater the difference between a signal's peak and average level, the more apparent the difference between RMS and peak compression/limiting will be. On a sustained pad sound with no peaks, there should be no appreciable difference. Peak sensing can sometimes sound over-controlled, unless the amount of compression used is slight. It's really down to personal choice, and all judgements should be based on listening tests.
