The current fall back is automatic, yes. I can see the exact percentage of people hitting it, which you won't so long as you can find a client to serve the files you are requesting.
When people actually report nodes that are misbehaving I take a look at them and manually reduce the cap on their load limits to a point where they aren't taking 5mins per page and they generally don't cause any issues after that point.
Most clients serve pages in 300-700ms per page - obviously for the 10mb+ pages it is a bit slower, but the majority of pages are reasonable size at under 1MB to a few MB at most - because I do keep an eye on stuff. Obviously it's impossible for me to test every single client myself accurately from my region so I do rely on reports from people in other regions to let me know if they are having an issue and of course we do have ways to automatically check on client stability and such, but just cos a client is serving people at 100% success rate, it doesn't mean that it is doing that at optimal speeds which we currently have no way to track.
As far as I recall, neither of you have ever actually reported a single bad client to me though, so you probably wouldn't know any of this and of course that means I've also never probably taken a look at any clients you are experiencing issues with, especially since you are both from different regions to me and thus hit clients I struggle to accurately get a read on. For example there are plenty of instances where a SEA client gives me 5 seconds or more load times, but people I ask in SEA to double check for me report 250ms load times per page.
Improvements to the network are planned however. There will be a way to report bad clients from the reader once it is switched over to v2 of the api, so you can report them from chapters. A "reload page from different client" button is on the to-do list, but is currently not possible without some additional updates. Port 443 only mode is currently available for testing on the beta site, so hopefully that will be added soon so that people behind firewalls they have no control over will be able to request only traffic from clients that can actually serve them.
