Speedtest.net is the most inaccurate speed test I've ever seen. I'll show you why:

With my test setup 150Mbit/s downstream / 20 Mbit/s upstream over a TL-SG108E I get following speedtest.net results:






Let's compare this with results from much more accurate dslreports.com speed test:






Now lets limit bandwidth in TL-SG108E to 10 Mbit/s ingress / 40 Mbit/s egress:






With speedtest.net I now get:






WTF?! Lets try dslreports.com speed test:





Much better, but still not accurate.

Now lets do a real speed test: using iperf running on a system connected to TL-SG108E's port 2 and its counterpart running on a server in the LAN:






As you can see, bandwidth limiting in TL-SG108E v2 just works fine (the initial spike in the diagram above is caused by the laptop's OS which buffers data and reports it to the iperf application as being sent while it actually has not been sent over the wire yet).

So if you want to use a web-based speed test, which might not always be accurate purely due to its nature (too much influence by lots of factors such as current traffic on their servers, method being used - keep away from Flash-based tests! -, ISP's traffic at the moment etc. etc.), use the speed test at dslreports.com. IMHO it's the most accurate and you will get some other interesting figures beside speed such as buffer bloat and ratings compared to average results for a given connection type (fiber, cable, DSL, WiFi) achieved by other people.

You're welcome. iperf is the standard method to measure throughput in TCP/IP networks. Current version is iperf3, see http://software.es.net/iperf/