I'm not sure these numbers are quite right, it looks like the author is just taking the instant values Power Gadget shows even when the clock and utilization are quite different at that particular moment - measured utilization is about a factor of two different. It would have been saner to dump the Power Gadget log and munge it a bit over some time window for more meaningful comparative data.

I should have planned it better and taken more before and after data with fan speeds. I was a bit worried if anything would have gone wrong this experiment would be an expensive mistake. Fortunately, everything was fine in the end. I have been using my laptop since I could see the difference. The Fans don't kick in that often and it had a noticeable impact on battery life.

Just guessing, but the fans might not start running that early with better heat dissipation.

Keeping the silicon cool reduces power consumption by about 3.5-4% per 10K.

But that should not be case for idle, so lower fan speeds (or no fan) seems logical

This is correct. Lower power consumption as a result of better TIM and thus lower core temperatures:

http://en.wikipedia.org/wiki/Poole–Frenkel_effect

https://forums.anandtech.com/threads/effect-of-temperature-o...

The effect is so pronounced that a watercooled computer running under full load will typically use less power than an identical aircooled computer, because the lowered silicon current leakage more than makes up for the power consumption of the pump and additional fans.

Maybe he's not the thermal paste genius he thinks he is and his paste job actually caused more heat, which decreased impedance?