Someone here remarked in the recent past that water-cooling may allow you to squeeze and extra 300 Mhz in processor speed from the system. That seemed like an observation grounded in experience.
Tom's Hardware Guide reviewed some 20 water-cooling kits a year ago, and the measured thermal resistance data on all of them was not impressive. I think the best of them showed a thermal resistance value of 0.20 C/W, which is higher than that of many heatpipe coolers now.
A top-of-the-line Swiftech kit achieves a thermal resistance of 0.125 C/W with sufficient fan-speed and throughput at the radiator. This barely "trumps" our best heatpipe coolers.
It may be that overall motherboard and CPU idle temperatures are lower with water-cooling, but if a heatpipe cooler shows corresponding temps just a few degrees Celsius higher and the idle-to-load temperature spread is the same or smaller, I just don't think it makes sense anymore, unless you're striving for noise reduction.
However, unless you grab one of those Reserator or "ThermalTake Rocket" kits, with the umbilicals from the external radiator to your computer, you still have to push air through a radiator, and you still have fan noise.
Water cooling requires periodic attention to fluid change, biocide and other features. Air-cooling requires the occasional application of canned-air for cleaning dust or "kruft" from the heatsink fins.
For my part, if I do it, it will be "chilled-water" cooling, because I just don't think water with a minimum temperature at the room-ambient is worth either the trouble or the expense. But that doesn't eclipse the 300 Mhz observation, either.