While the exercice is interesting in that it is humbling for non-native speakers, it does seem dubious. I repeated the test 3 times, and got widely different results (10400, then 16700, then 18300). Since they are all quite outside the reported quantiles for native speakers, it may be expected to have high variance for non-native, but that means the numbers are not very useful as is.
I can't confirm your observation: my first attempt got me 11,500 words, the second 11,800. Of course I didn't check the words that I learned from the first run (I used a dictionary to verify whether my understanding was correct when in doubt, and to find out what some of the words mean that I didn't know).