I believe the most significant factor why the OOKLA-based tests show a higher reading is because they create a swarm of multiple connections. If the network is at full capacity or there is excessive latency for another reason, the extra connections can push more data through. Usually they together compete 'as equals' with other data streams active at the same time.
In this case I got 7 connections to SpeakEasy at Washington DC. The number seems to be variable, and, as far as I can see, I cannot influence it.
Of course, if everybody opened more connections to get higher speed, nobody would get it. A somewhat related discussion about network congestion and the FTP protocol.
I think a more realistic test is a download of a controllable number of files from a server or servers with known capacity much higher than your line is. That is what you could do in normal use as well. Starting 7 downloads and fragmenting your disk? Probably not. Using special software to spam the network? Probably not the right thing to do.
LeaseWeb provides a good, unbiased download speed test with low CPU overhead (unlike Flash tests). I do not know a good public place to upload to. If you have an FTP server, it will work fine, because you can easily start as many files as you decide.