Like anything else, the speed test systems are subject to congestion, routing/peering and loading issues; I guess that's why the main BT test site limits retests to take place after an hour, and even then it's not great.
http://speed.io is an alternative but still subject to the same kind of limitations, and the
BT beta test for >40 Mbit/s often seems to be able to drive things flat-out (but perhaps not for much longer

).
Some services - like the news feed you mention - often seem to have more capacity than the testing services; e.g., I've seen iTunes run flat-out at 4.6 MB/s for >10 mins at a time. This is hardly surprising given the job they're intended for, whereas most speed test systems seem to be a "best efforts" affair.
Also, don't always rely on the result given by by the testers as they may average rise/fall and drop-outs. Use network monitoring tools(*) where available to monitor "instantaneous" (really short-term averages, e.g., 1s) rates and average over the flat-out values, since peaks may be higher due to buffering.
Although it's not so useful for bandwidth checks,
Netalyzr at UC Berkeley is useful for a general health check and spotting - if not fixing - buffering and latency issues in your ISP's network (wired ethernet please, unless you want to see the consequences of running wireless).
* I think TBB have something for Windows; there's Activity Monitor on Mac OS X and there are various things for *N*X systems, e.g., bwm-ng, ntop etc.
Edited by deleted (Fri 03-Feb-12 11:43:08)