Hello solope:
When you said you "reset the CMS" I'm assuming you meant that you reset the SMC?
Anyway, I wanted to report my experiences today which may shed more light on the issue. I reset my SMC for the first time late last night and then ran another "work day" battery life test during the day today. My work day involves running a Parallels VM with W7 and maybe 10 windows going in it--some compiling, use of Eclipse, typical developer environment.
Shock of all shocks, whereas the day before I only managed 3.5 hours with the VM going, today the machine gave me much more. I recharged when I reached 8% level, after 06:40 running and with 00:37 estimated remaining. So it went from 3.5 the day before to over 7 hours today. Now, if someone had told me I'd be looking at "only" 7 hours with a full day of Parallels VM with active development, I'd say that's pretty good--maybe even better than I would have expected given the heavy energy use that a VM typically is. Besides resetting the SMC, I did make a couple of other changes: I put the display brightness down to about the minimum I could handle (from 50% the day before to 25%) and also turned the keyboard backlight from 1 bar (minimum) to completely off. Like the day before, I was plugged into a powered USB HUB because of various hardware things I'm working with for development (and Time Machine was running hourly). This raises the possibility, I guess, that the HUB could be providing some power to offset the machine battery drain. But I certainly didn't see things going that way yesterday with the exact same connections. So I'm guessing not on that count. (Don't have a clue if a self-powered USB 3 HUB could provide any power into a USB-C connector from the MBP.)
But resetting the SMC seems to be the main element that would explain the improvement. Take now, for instance, I'm unplugged in complete stand-alone operation back with 50% brightness and 1 bar (minimum) display backlight and the battery indication, after half an hour has steadily RISEN from startup at 100% charge--where it showed something alarming like about 4 hours, to now indicating about 8 hours estimated remaining. And it looks like it will continue to rise further still. (This is probably because the energy use at initial startup is artificially high and skews the digital filter which smooths out the prediction.)
We'll see what happens this evening as I continue to use it without the VM, but initial indications are that I'll see way better than the 4.5 I saw in this same situation yesterday. (Yesterday, I NEVER saw the estimate rise. It started out pretty lame and stuck there or degraded a bit.) So something significant has changed.
I have no knowledge of the specifies of the SMC, but I did migrate my settings from a previous 2011 MBP time machine backup right after getting on this new 2016 MBP. Some suggest that the migration brings over information from the previous machine which could confuse the estimation by the new machine. That seems strange that they would allow hardware-specific energy data to come across during a migration, but I have no idea. Nor would it explain what you seem to be seeing having gone back to clean installs.
About all one can conclude for sure is that it is worrisome and more than confusing. ;-P
(Yeah, in the time it has taken me to type the previous two paragraphs, my battery prediction is now estimating almost 9 hours--I've gained an entire hour based on my present energy usage, apparently, in about 10 minutes or less. Overall, this is a potentially positive outcome though!)