Atypical language acquisition (e.g., as a second language, or through a non-standard channel like technology or fantasy literature) disrupts this extrapolation step. For instance, a German programmer that knows the word "polymorphic" via OOP is less likely to know similarly frequent and difficult but programming-unrelated words than British or American peers. So adding, say, 100 to the total would be utterly unfounded. Same thing for a science-fiction nerd: Acquaintance with obscure words from one domain doesn't extrapolate to other domains.
Unless they somehow control for domain specificity and atypical acquisition, let's not get too frustrated. (Disclaimer: Not a native speaker -- result around median.)