Appears they have had to take actions in filtering, shaping and as well QOS + deeper, would be my first impression. In the attempt to spread what throughput is available among requests. If this was not implemented, as you know, one node could easily absorb all available bandwidth, leaving others to suffer 56k experiences.
They are most likely doing the best they can with what they have, in order to bring revenue to expand throughput not necessarily bandwidth itself.
Though in theory what they have sounds infantile. Since it does not function as intended. If so, at the load factor on the node, no one connection could exceed the spec, your basically bypassing their config by P2P, which explains the majority are not utilizing the protocol. Hence, they could consider you part of the problem lol. Which of course you are not, though in terms of allowing all customers to have access at a specified minimum level. [at peak levels]