If the metrics turn out to be correct, I think the most interesting thing is that the cache hit rate seems relatively unaffected. Which would suggest that we could clean-sweep the cache on restart without penalty.

I moved my chromebook over to beta to check this out since the metrics aren't on stable yet. I see three SimpleCache.Http.PercentFullOnInit events, two of size 0 and 1 that's full. So ChromeOS must have some extra caches it uses for something else. So WAI I guess. I wonder what those caches are though.

Can we move the metric to only report HTTP cache size?

That's exactly what it does. I'll need to dig into the chromeos code to figure out where the extras are coming from. I was going to make a new bug for that but it makes sense to continue it here.