The evidence is definitely beginning to show that it is indeed true that end-to-end throughput between arbitrary end-points on the Internet is indeed slowing.
About a year and a half ago I found some postings by Jim Gettys (one of the co-creators of the X Window system) that described a similar problem, but went on to explain why this is happening. He even coined a phrase for the underlying problem: Bufferbloat. See »gettys.wordpress.com/2010/12/03/···erbloat/
To oversimplify the problem: Equipment vendors focus on moving your data from where you are to the next hop (i.e. your Cable headend, DSLAM, cell tower, etc.) as fast as possible, without regard to the nature of the destination of that data. Now that we have cheaper, faster, bigger CPUs and memory, it has become trivial to do this.
After a quick Google search, I did not find a nice overview article, but if you Google for bufferbloat, you will find plenty of material.
Larry Gadallah, VE6VQ/W7
616D 4E52 CF1F 3FEC FFFB
F11B 7DB9 C79A EA7E B25B