He claims the core problem is congestion collapse caused by buffers that aren't configured properly. Wikipedia says when this happens, traffic demand is high but little or no useful data can get through due to congestion.
Mr. Turner gives his synopsis of what he thinks is going on based on conversation threads he has been following and the results of some tests he and others have run. You can read the whole article in its technical glory for more information, but the bottom line is it appears AT&T has configured their network for zero packet loss. He seems to be arguing that packet loss is better than buffers configured to hold a ton of packets so there is no loss. TCP networks have mechanisms to recover lost packets, so no data would be lost. It would apparently take less time to recover the packets than to have the network configured so they are never lost in the first place.
Whether or not the problem has been properly diagnosed, I have no idea, but it is interesting to think that many of the issues that have been suffered by AT&T subscribers could be fixed by some tweaks to how the carrier's network operates.