Unless activity for any given IP reaches epidemic proportion I ignore (eg) ThePlanet, AWS etc - in fact, any server farm I discover. :)
I have large chunks of AWS in the IIS firewall so do not see stuff within those ranges. Other stuff seems to be spread pretty much across the world, although much of it from USA, Russia and Ukraine with a fairly high proportion from China. And, for some reason, "ain't I clever" UK idiots trying it on for no obvious motivation other than self-gratification.
I'm working on something like this right now, allowing you to associate user agents with IP ranges. I've been collecting data since 2009. I have one site which gets hit pretty hard with scrapers (as well as legitimate SE bots) so it's almost like a honeypot for finding new and obscure user-agents.
I'm hoping to work the data in such a way that it will be possible to determine, based on past empirical data of activity on various sites, whether a load from a particular IP range is likely to be a bot. As the OP hints at, you wouldn't expect many interactive browser sessions to be coming from AWS or ThePlanet IPs, unless the servers were proxying for humans... I'm going to be trying to determine that automatically.