I am not trying to or have any desire to ....shoot down what you have done (well perhaps a little bit...you have certainly made at least one glaring errror.)but to produce suspicious data without showing or describing your test setup is definately NOT going to win you any friends.
A CPU is not a regular heat source even under stress test and certainly not with runs of tubing between them TECs change temp in the blink of an eye even a fractional drop in the temp of the coolant from the CPU will produce an instantaneous change in the TEC which will take sometime to even off again. A change in the load can change the current draw, the Dt and finally the hotside temp which changes all the characteristics of the TEC. The charactertistics of a TEC depend on a constant hotside temp. This is how charts are produced.
How were the TECs setup within the loop, how did test temps.
Also could you identify the TEC you are using please to say 245w is totally insufficient.
You have added the wattage from the wall to your supposed coldside load ???
The load on the coldside is only your supposed 160w CPU plus any dump from the pump and if you set it up properly would of been the same all through testing.
Wattage from the wall plus the heat dump is just what you can expect to be ejected from the HOTSIDE and therefore would not seem to be part of your experiment, except perhaps to just see a rise in hotside temp. What did you use to measure that ?
The change in temp on both the hotside and the coldside just dont figure, I am really curious how you arrived at those temps. Certainly the coldside temp whilst testing is one of the biggest PITA's going I really fascinated at how you measured that.