I think it's has to be a combination of really sound design from the ground up, done by people who really know their stuff, plus fullest usage of QoS at all stages as I can't see it having a chance without. And as I said earlier I think there would absolutely have to be L2- as well as L3-QoS otherwise there would be no chance of it working half the time, as you could easily get a very congested home wireless LAN, never mind the internet access pipe and wider internet, and it could easily completely fall apart just because of the WLAN so that means L2-QoS ("WME" / "WMM").
Andrews and Arnold's VoIP as far as I know (but what do I know) doesn't use L3 QOS at all, so whether it works or not relies on the power of prayer. The ISP-end Firebrick routers prioritise small packets as a heuristic in the hope this will work, but this could fail in a pathological case where there is congestion elsewhere or if there is competing traffic from an odd alien small-packet protocol running flat out. And in the case of upstream, then the user could be using a random router so in that case AA has no chance because they are not in control. I would like to see the VoIP system changed to use L3 QoS everywhere. Firebricks at the home user end should also use L2 QoS marking and remarking in the direction L3->L2, and possibly the reverse too L2->L3 but perhaps only if the packets have no L3 markings or some such rule, possibly configurable. I can think of more fancy mapping schemes too. There is no mention of the use of QoS anywhere, the docs just mention small packets, so I may well have got this all completely wrong, in which case apologies to the Firebrick devs.