Would an ISP still be able to call their service "unlimited", by this definition, if some of their distribution nodes (cable optical nodes, wireless cell towers, whatever) are oversubscribed+saturated to the point of degraded service for the customers on that node?
If they could, then would they still be able to call the service "unlimited" if they enforced moment-by-moment QoS at the distribution node (so your neighbour's download will slow down to allow you to Skype, as if you were both connected through the same router in a building)?
And if that is allowed, then what is the difference if they're doing the QoSing by long-time-window statistical aggregation rather than packet-by-packet prioritization?
Decreasing the bandwidth available to a given link that has reached some monthly "quota" on a given oversubscribed node†, has exactly the same incentives as decreasing the bandwidth available to a given link for as long as someone else is trying to saturate the pipe with higher-priority traffic.
Either method disincentivizes everyone on the link from "wasting" their bandwidth unnecessarily (in a "you don't want your neighbour to do X, so you stop doing X and hope that they have empathy" way), such that the link becomes less saturated on a long-term basis. But only one is ever highlighted by the news as explicitly breaking the promise of "unlimited." Why is this?
---
† (...where that quota is the area-under-the-curve of the router's upstream bandwidth, divided by the number of users attached to the distribution point—in other words, the user's ration of the distribution node's available transit. If the quota is an arbitrary fixed number that isn't anywhere near this number, then that's just price-gouging.)