\cc @tstarling

We talked about this today and I mentioned remembering that there was an issue with this churning the cache on plain installs, and that I was wondering at the time whether we still needed this additional layer.

In context of change 771762, the answer is yes we do still have a use for quick touch, which is to specifically avoid 304 responses when the user has recently recieved a notification. For that purpose, the lightweight memcached layer is ideally suited.

Having said that, I do think we can probably improve this to not blindly initialise User::mQuickTouched for every user on every page view. Perhaps we can make it smarter and give it a bottom value as default that we don't need to store but still consider into HTTP revalidation, and then only set/bump it when needed, which should still have the same effect.