Your description of the symptom was more or less the same with my personal experience, but the conclusion might be a little bit off IMO. I first tried to deploy proxies to google app engine and use PAC scripts to auto switch between them. It was super fast and worked for ONE day. I really meant one day here, the whole thing stopped working the next day obviously the maneuver was detected and busted. Then I tried both public and commercial VPN services but all of which were either unstable or slow, or even both. At the same time, I almost never had any problems with corporate VPNs but I did not want personal stuff to go through company networks so I hardly used them unless had no choice. Things did not change much until AWS/Azure emerged. I almost immediately deployed an Azure VM dedicated for VPN when it's available to my MSDN subscription. I was pretty comfortable with it as even if my VM got blocked, I could actually redeploy it to a different region in minutes. It turned out I worried too much, I never had any drama with it since I kept very low profile and the VPN was really for myself only.

DNS pollution was for sure only one means of the blocking. Even if the IP address of the site got resolved correctly, the site could not reached as a result of "Connection reset" or "Remote host closed the connection" errors. I thought the blocking was all the time and everywhere until I accidentally realized that I had just access some site without VPN connected and such state could last for from hours to days, as long as I did not access contents deemed sensitive, sites did not seem to matter that much as blocking would not be triggered until I clicked some links. HTTPS did not seem to help at all, so GFW must have the ability to do deep packet analysis. Such behaviors make sense to me as the network traffics in China are enormous even the government would not have enough resource to monitor everything all the time, so the practical approach would be using a little bit of heuristics and commencing blocking only when certain signals were triggered. Also I encountered a couple of times man-in-the-middle attacks as I noticed my browser were not happy with the site's certificates. Such attack might be carried out by the ISP as the certificates were self signed.