I want to commend @AlexR for trying to raise the profile of the bufferbloat problem with manufacturers as well as their customers.
@Weaver - you're asking the right question: how can vendors still claim to have good implementations relying on a single queue for all traffic?
Bufferbloat bites you at the bottleneck link: usually that's your connection to your ISP. It'll happen in either direction, down or up, depending on the gear at the ends. Using fq_codel, you can take control of the bandwidth in *both* directions, so that you minimize bloat/latency each way. I would recommend that everyone try the speed test at
www.dslresports.com/speedtest since it measures the latency *during* the download and upload. (I suppose that it's possible that your ISP's internal network is more bloated than the link to your home, but that is not really likely long-term.)
I've been following the discussion on the Bufferbloat.net site for about three years, and it has been delightful to see the improvements in latency/response time after implementing fq_codel.
In fact, I'm going to put my money where my mouth is. I plan to record a podcast next week over Skype while using netperf to fully-load my 7mbps/768kbps DSL circuit. In practice sessions, it has worked absolutely fine - no garbling, dropouts, etc. as long as I use the fq_codel in the OpenWrt firmware. If it turn it off, the Skype session gets as bad as you might expect.