I've never gotten useful results from powertop. The model it builds to correlate system events vs. observed power usage always diverges with reality very quickly. In the good case it'll just do something a little silly like conclude that the CPU fan is the ultimate source of all power consumption. In the bad case it'll be something like think that a pair of linux veths uses 3kW (not a typo) of power.
I was the same - mixed results at best. I figured that manually playing with this all the time was the wrong solution and installed TLP instead which seems to work pretty nicely. Not sure if it is mentioned in the linked article, so if not then check it here: http://linrunner.de/en/tlp/docs/tlp-linux-advanced-power-man...
