I'd say it is more of a problem of incorrectly configured QoS, or hardware with insufficient QoS capabilities, rather than large buffers. Obviously they are not using WRED or other methods, or the thresholds per queue are set too high to activate WRED or other packet drop mechanisms. This results in the buffers always being near 100% full, during periods of congestion. There are a slew of QoS capabilities on different hardware from different manufacturers, and even from the same manufacturer. Cisco, for example, has different QoS capabilities on almost every different piece of hardware they sell. So, you have to be fairly diligent that you are configuring QoS correctly on each individual piece of equipment, many of which will have very different capabilities, to be able to ensure an overall QoS strategy for the whole network.
However, this proper functioning of QoS is, as anyone who really knows QoS, dependent on the proper configuration on every node in the network. If you are talking VoIP, for instance, just one improperly configured node, or even a single link on a node, can break QoS on the entire network (or at least flows going through that node/link). Since most cheap home equipment does not have configurable QoS settings, or at least not to the extend that Internet infrastructure devices do, they may well be part of the problem.
However, as far as the Internet infrastructure devices, if Comcast, or any other ISP, is suffering from "buffer-bloat" on their equipment I'd blame them for not configuring QoS appropriately.