I don't claim to know the technical details about TEC's, but my experience with them has shown that compressor-based chilling is just more efficient.

TEC's and their inefficiency seems to be a common thought 'round the 'net:

http://www.coolchips.gi/technology/ccalc.shtml
For comparison purposes, a typical thermoelectric (Peltier) cooler has a Carnot efficiency of about 5-8%. A typical domestic refrigerator compressor has a maximum Carnot efficiency of about 45%.
Pelt Basics:
http://www.overclockers.com/tips45/

http://en.wikipedia.org/wiki/Thermoelectric_cooling
Thermoelectric junctions are generally only around 5–10% as efficient as the ideal refrigerator (Carnot cycle), compared with 40–60% achieved by conventional compression cycle systems (reverse Rankine systems like a compressor). Due to the relatively low efficiency, thermoelectric cooling is generally only used in environments where the solid state nature (no moving parts, maintenance-free) outweighs pure efficiency.
How you got those results, I don't know. Due to the Laws of Thermodynamics, I do know that 8 watts of heat movement + 12-18 watts of heat source != more than 19 watts of heat movement total.

Either the temperature measurement was off, the TEC was improperly rated, or there was something else helping cool the water.