All offline page efforts rely on MHTML snapshots to store web page contents. Beyond other benefits, there's the belief that it is a faster alternative compared to re-loading the contents from the web, especially in EM markets. But this belief has not experimental data to back it up and we should fix that.

We should add metrics to help up assess what's the actual performance of loading a page from a MHTML file. Some suggestions of what to track (each item might relate to more than one histogram):
* Time to load and parse a MHTML file (MHTMLArchive::Create and maybe other places in its call-chain).
* Loading and parsing bandwidth: time-to-load/file-size.
* Time to fully load a page from a MHTML file, from navigation request to load finished.
* Full navigation bandwidth (maybe useful?).