With TestMy.net you are testing to Dallas TX, other test locations are available around the world but none in South America. I'm still looking for a fast and decently priced server in South America.
There are many reasons why speedtest.net results differ from TestMy.net. Here's a couple of topics that will shed light.
Why Do My Results Differ From Speedtest.net / Ookla Speed Tests?
Internet providers caught inflating speed test results
You said it yourself, you're getting crappy quality from netflix while speedtest.net says you have 20 Mbps. Sounds like TestMy.net is more representative of what you're really seeing. You're not alone, topics like yours are by far the most common question we get here.
You already have your answer and know which test is right. Why would you trust that test when you can feel that it's much higher than it should be?
I have an option that you can use to make your own multithread download tests against pretty much any server you want. I created one for you using netflix servers.
:::.. Download Speed Test Result Details ..:::
Download Connection Speed:: 98183 Kbps or 98.2 Mbps
Download Speed Test Size:: 96 MB or 98328 kB or 100688125 bytes
Download Binary File Transfer Speed:: 12273 kB/s or 12.3 MB/s
Tested At:: http://TestMy.net Version 15
Client Stats:: https://testmy.net/quickstats/CA3LE https://testmy.net/compID/4602014672148
Test Time:: 2016-02-02 07:27:42 Local Time
Client Location:: Monument, CO US https://testmy.net/city/monument_co
Target:: assets.nflxext.com https://testmy.net/mX/KsSMa
Client Host:: Comcast Cable https://testmy.net/hoststats/comcast_cable
Compare:: 29% slower than client avg, 137% faster than host avg, 69% faster than city avg, 217% faster than country avg, 531% faster than world index
1MB Download in 0.08 Seconds - 1GB Download in ~82 Seconds - 1753X faster than 56K
This test of exactly 98328 kB took 8.208 seconds to complete
User Agent:: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_11_2) AppleWebKit/601.3.9 (KHTML, like Gecko) Version/9.0.2 Safari/601.3.9 [!]
Comparing to multithread on TMN Dallas where I got 135 Mbps using the same test size immediately after. And then a linear result...
This, by the way, is on wifi that I know to be under performing. On a computer directly wired to the router with cat-5 moments later...
The best thing you can do to understand your connection and where it's weak is to test it more with TMN. By altering the variables, testing against different servers, using the different test types you can spot weak points in your route.
Hell, you can even use TMN to test against speedtest.net servers if you'd like. Search google for "speedtest.net server list" and you'll instantly find their publicly available server list. Search the XML file for your city or the server you know about already. Take the URL e.g. "http://speedtest.com.uy/speedtest/upload.php" then drop the "upload.php". Most of the servers are open to file navigation, just open the URL in your browser and you'll see the list of files that speedtest.net uses to test. Copy the URL for the packet size you'd like to test with and paste it into http://testmercury.net ... TMN will turn the image into a speedtest that can be automatically sized anywhere from 2X the original size to 200 MB. Literally just paste a URL to a valid image online and hit enter...
So here's the directory listing for a random speedtest.net server in UY
That's the standard file structure for speedtest.net servers. You could just take "random350x350.jpg" and add it to the end of any of the server addresses on the XML and it will make a great test using TMN instead of speedtest.net. Faster connections you may want to use the larger images, which are then repeated to reach the final test size.
Hope this helps you nail down the problem. Always compare the multithread results (the Mercury test is multithread) to the classic linear results.