I realized that Memory.Renderer.Large2 is broken from Android M and above.
Android M introduced some SELinux policy that prevents the browser from reading the /proc folder of renderer. I cannot trace back what change this is. On doing "setenforce 0" the metrics are recorded correctly. I think this is because my local build is an untrusted app. But lot of phones do not record Memory.Renderer.
UMA links: 
Nexus 5 
https://uma.googleplex.com/p/chrome/histograms/?endDate=20170319&dayCount=1&histograms=Memory.Browser.Large2%2CMemory.Renderer.Large2&fixupData=true&showMax=true&filters=channel%2Ceq%2C4%2Cplatform%2Ceq%2CA%2Cosflavor%2Ceq%2CAndroid_Marshmallow%2Chw_class%2Cone_of%2CNEXUS%205%7CNexus%205%7CNexus%205%20CAF%7CNexus%205%20SkyDragon%20Rom%7CNexus%205%20SkyDragonRom%7Cnexus%205%7CGoogle%20Nexus%205%7CLG%20NEXUS%205%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial
M Nexus 6P
https://uma.googleplex.com/p/chrome/histograms/?endDate=20170319&dayCount=14&histograms=Memory.Browser.Large2%2CMemory.Renderer.Large2&fixupData=true&showMax=true&filters=channel%2Ceq%2C4%2Cplatform%2Ceq%2CA%2Cosflavor%2Ceq%2CAndroid_Marshmallow%2Chw_class%2Ceq%2CNexus%206P%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial
M Samsung M (it works).
https://uma.googleplex.com/p/chrome/histograms/?endDate=20170319&dayCount=14&histograms=Memory.Browser.Large2%2CMemory.Renderer.Large2&fixupData=true&showMax=true&filters=channel%2Ceq%2C4%2Cplatform%2Ceq%2CA%2Cosflavor%2Ceq%2CAndroid_Marshmallow%2Chw_class%2Cregex%2CSAMSUNG%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial
My belief: I always thought browser cannot read other processes /proc directory because of uid isolation. But magically it works on Samsung?

On Android N I think this restriction got stricter and even the trusted apps cannot read the renderer /proc files. This leaves us with NO phones recording Memory.Renderer metrics in N and O.
UMA links:
All N
https://uma.googleplex.com/p/chrome/histograms/?endDate=20170319&dayCount=14&histograms=Memory.Browser.Large2%2CMemory.Renderer.Large2&fixupData=true&showMax=true&filters=channel%2Ceq%2C4%2Cplatform%2Ceq%2CA%2Cosflavor%2Ceq%2CAndroid_N%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial
Samsung N
https://uma.googleplex.com/p/chrome/histograms/?endDate=20170319&dayCount=14&histograms=Memory.Browser.Large2%2CMemory.Renderer.Large2&fixupData=true&showMax=true&filters=channel%2Ceq%2C4%2Cplatform%2Ceq%2CA%2Cosflavor%2Ceq%2CAndroid_N%2Chw_class%2Cregex%2CSAMSUNG%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial

Given the facts above Memory.Total is pretty much garbage since it records just browser and gpu on some cases and renderers on some cases.

Related bugs and discussions: 
issue 581517
issue 581517

Do we wait till Memory-Infra UMA to fix this?

Primiano:
We solved this problem back then in memory-infra ( crbug.com/461788 ) by having logic that, on Android, dumps these stats from within each process, but we don't have something similar in the current UMA.
Given the current ongoing work, not sure how prioritary is to fix the existing UMA memory reporting pipeline, given that we have been in this state for a while now (however, if whoever maintains that wants to go ahead and fix it, I don't have any objection)