Wireline QoE degradation: further explanations (maybe) ?

Last week I posted on the sharp dip in Akamai's average speed data in many countries in the world in the last quarter or two quarters. The piece has gotten some attention and a few responses were sent out to me. Five in particular suggest possible explanations as to why the data might be going down other than the core hypothesis (the strain of growing demand is finally outpacing network upgrades). None of them necessarily explain the global data dip, but they are all interesting in their own right:
  • Some ISPs may be throttling traffic to CDNs, as does Plus.net in the UK. I find that hard to believe on a global scale (in other words, some may do it, but we'd need most of them to do it to explain the dip).
  • The multiplication of wireless devices hogging the home wi-fi may explain a saturation before the connection even hits the wireline access line. That's a well-known phenomenon and could explain some of the dip although in western markets WiFi is now at a capacity of 54Mbps or more, much higher than most access lines.
  • Netflix has officialised today the fact that they'd been building their own CDN for traffic. I'm not sure how that could affect the data (the reader who suggested this wasn't explicit and I can't quite figure it out), and if it did it would only be for North America.
  • Someone suggested that Megaupload's demise could be the changing factor that affects the data globally. I find that one intriguing because indeed there are few discreet factors that could explain a global change in average speeds and Megupload is such a global shift. However, I can't reconcile cause and effect: if a massive drain on the network suddenly dissapears, surely speeds should increase not decrease. Unless Megaupload traffic has been substituted by another kind of traffic (streaming, P2P…) that affects network performance negatively…
  • In some markets like the UK, superfast offers are being sold on the access, but backhaul capacity may not be upgraded accordingly, hence a crunch in performance for most users. That is possible indeed but wouldn't affect the data globally. 

Meanwhile, I haven't been able to clarify the likelihood of the AP hypothesis expressed in the original piece, so if anyone has views on that, please share them!