I used plain NLTK token analysis on rap genius lyrics. in terms of several more data slices...I agree that there should be more cuts of the data, but you must understand the amount of time that it took me to put this together.
You see, I have a strong opinion on that any data-analytic work is pretty close to being useless if it's not reproducible. And I mean really close. I already mentioned few questions that naturally arise reading your article which are crucial to understand your results and are not addressed by the article. So, ideally, any data analysis done for the open community should provide both full dataset and full sources. Unfortunately, it's not always possible: dataset may be several terabytes long, there might be legal issues about disclosure of the data (or letting everyone know how exactly author acquired them), sources might be under NDA, whatever. In this case description should be as pedantic as possible, because some little details can change the whole meaning of statistics entirely.
By the way, NLTK allows you to do all kinds of processing, so it isn't the answer.
So what would I do? The usual course, actually: would do the work in iPython Notebook, cleaning it up afterwards, would've drawn graphs in place and printed some few slices I've said already while processing, so it would be easier to understand what actually that unique words counted might be like. While fancy d3 graphs are cool for sure, but not nearly as useful.
While Aeson Rock and DMX are clearly extreme and not surprising at all, it's not that clear for some guys in the middle.
So, first off, for every data project sources should be provided, or at least more specific definition, how text was processed, tokenized, analyzed. Second, several more "data slices" should be provided, for instance 100 most used words which are unique for that artist compared to other artist in the list.