It would be interesting to know how much of that 3C is the increase in airflow.
If you would be interested, some experiments with your system could yield some valuable information for the TEC simulation effort being discussed
in this thread.
What I had in mind, was something like the following:
1. Move the radiator fan to a fixed 12V supply. (This would keep the water cooling at a constant C/W)
2. Set the processor clocking to the factory settings.
3. Set the TEC supply at 15V
4. Let the processor run loaded for an hour.
5. Record every temperature you have access to. (for example; ambient, on die, socket thermistor, MOBO, water temp, etc.)
6. Repeat steps 4 & 5 except with the voltage dropped 1V. Continue repeating the 1V drop until the CPU temp is 40C+. (Or as high as you are willing to go.)
This data would allow me to calculate a ballpark number for "Rins" which represents the heat that leaks through the processor pins and other heat leakage paths. Lack of a good value for Rins is one of the biggest issues in the way of being able to calculate pelt CPU cooling accurately.
This would be a fair amount of work to do right, but if you'd be willing to do it, let me know.