I haven't looked into it yet (whether the benchmarks are comparable) but it would be interesting to check the programs there against the ones in this post, and either update the benchmarks here or add new submissions to that question.
For a while the fastest program there was one in Rust, then out of curiosity I translated Knuth's program from 1986 into C++ and a simplified version based on it was even faster; currently a Rust translation of the idea (a custom trie using not too much memory) is the fastest one there. (In fact, the current fastest two Rust submissions there came after I linked to the question from HN in Feb 2020, in a couple of comments at https://news.ycombinator.com/item?id=22221592 and in the discussion on a "Donald Knuth was framed" article: https://news.ycombinator.com/item?id=22406070.)
Edit: I tried to do the comparison myself but one issue I ran into (apart from minor things like taking input from filename vs stdin, and printing top N words versus all, which are easily handled) is that the benchmark on the StackExchange question considers "words" to be made of alphabetic characters (a–z), and the trie solutions make use of that fact (a trie node has 26 children etc), while the benchmark from this post includes punctuation (e.g. it has "words" like "him," and "him." i.e. "him" followed by a comma or period are counted as separate words). I gave up at this point, but maybe someone with more energy could modify the benchmark here and try comparing them.