You're citing information from two different studies that could very well have used different methods for counting lemmas. The text in between the two bits you quoted illustrates the extent to which this is the case: Alongside your study of 20-year-olds finding that they recognize ~42,000 lemmas, there's an earlier study of people about the same age finding that they recognize between 12,000 and 17,000.
The magnitude of that difference (250-300%!) should serve as a stark illustration of just how hazy these estimates are, even within a language.
Between languages, it gets even more difficult. It's very hard, possibly impossible, to come up with a single standard for what counts as a word that doesn't favor one language over another. For example, I wouldn't be surprised if you could engineer a comparison that says that English has more words, and also one that says that German does, simply by changing how you deal with agglutination. Or, if not that, then certainly agglutination and polysemy together.