Just what would you expect "A" or anyone else (including me) to do in such a severe bandwidth oversale situation that traffic is basically being throttled into oblivion? No amount of adjusting to the protocol I'm using on my server could ever solve this. At some point, the packet loss probability gets so high that there's a high chance that the TCP retry counter gets exhausted before the packet gets acknowledged and you can't even login. And even if you can login, if you are using TCP instead of UDP, TCP's congestion control kicks in and reduces the total amount of data that can be sent without acknowledgement. If you're using UDP, UDP doesn't do congestion control but TCP traffic sent over a UDP VPN or congestion control at the application layer will do exactly the same thing.

In any case, I work on the principle of "Never attribute to malice that which is adequately explained by incompetence"...or profit-seeking, as the case may be. My current theory is that they're using some sort of algorithm at the routing on international border that prioritizes certain types of traffic. Among the types of traffic with priority would be:
1. Encapsulated telephony traffic (i.e international roaming SIM cards that tunnel data back to their home networks)
2. HTTPS connections initiated from outside China to servers inside of China (i.e people outside of China purchasing stuff from Chinese companies -- because those companies pay for priority since lagging connections can mean a loss of business for them)
3. Anyone who pays additional money for data priority during congestion

I suspect that either the system looks at total amount of traffic load going over the international boundary every 10, 15, 20, or 30 minutes and adjusts priority based on the traffic load to prevent the link from saturating, or the ISP has a fixed allocation of bandwidth and basically does the same thing. This would explain the sudden changes in QoS -- VPNs working one minute and then failing to work for about 20-30 minutes and then working again. It's notable that this type of algorithm would produce this type of oscillation in QoS, as de-prioritizing certain people's traffic would cause them to produce less of it, reducing the total load on the link and then causing it to lift the throttling at the next iteration. Or they apply some sort of "round robin" de-prioritizing of different customers, kind of like a rolling blackout for internet service.

You know how Chinese ISPs/mobile phone plans tend to be ridiculously cheap? I think this is just a consequence of that. Driving around Beijing with my Unicom HK card, I would frequently see dropouts in usable service at certain times of the day, while at other times, service would work perfectly fine in the exact same areas. Setting prices extremely low and overselling service has consequences. There isn't really an issue with being sent to the back of the line if there isn't a line, but when there is a long line, being sent to the back is painful. Add to that the fact that people need to use VPNs in the first place and international internet traffic becomes extremely impacted -- even requests to internal Chinese websites, which shouldn't use any bandwidth on the international links, now consume double the amount of bandwidth over those same links.

So if my wife's parents have some sort of "el cheapo" internet plan that has a very low priority for cross border traffic, it's possible that they just never noticed how bad it is. They don't read English and so they pretty much only use Chinese websites, and again, being at the back of the line doesn't mean anything if there isn't a line. I told my wife to give the ISP a call but her parents were already sleeping so she'd have to get the account number from them first.