T148018: Use fetchStart instead of navigationStart for Navigation Timing to measure the actual page performance touched on this before, but wrongly assumed that we used navigationStart at the moment. We don't for all browsers, we use startLoadTime for Chrome.
The investigation in T100999: Make the logo's loading priority higher reveals that there is truth to the fact that fetchStart should be taken into account.
Moreover, the use of startLoadTime doesn't always work, which probably explains some odd values we were getting for firstPaint before (huge negative integers), as demonstrated here: https://jsfiddle.net/9fufvLyt/1/ It might be due to the iframe, but it's noteworthy. Direct link to destination: https://dl.dropboxusercontent.com/u/109867/test/index.html
We need to review our calculation of firstPaint in Chrome, as it might be wrong and lump things into that metric it shouldn't. It might mean a breaking change in regards to our historical data, but so far it seems like Chrome was inconsistent with other browsers, which means that our existing data is already polluted when aggregating multiple browsers.