Stylistic consistency is very good when you know what you're looking for, and the universe you're looking into is also comparatively limited.

It's been used for more than a century on literary scholarship to attribute (and un-attribute) works to authors. With the advent of computers, it is made a lot easier. You can introduce many more data and process them much faster.

[Even so, a known drawback is that styles change through time. Ideally, you would need to have works spanning all of an author's productive life. Some other things also produce interesting interference effects which may be confusing. Still, it works, and it is true science. It's just not accurate to the degree of a DNA test.]

However...

You would need a confirmed sample of text by "rocinante" (the bigger the better, of course), and then the sample text you are testing to see if it matches. If it is just a small sample: a specific book or a series of articles, letters or blog posts; then it's easy. But to crawl the whole web to look for a match is not very realistic. I suppose the computing power is probably attainable in the current technology. Nevertheless, it would seem to be not the best usage of this technology, when it would likely be more useful to detect truly dangerous criminals and terrorists than "just" Pedos. The text to be crawled is just too much for it to be worth it.




Disclaimer: I have experience in this sort of text analysis. I've also previously applied it on GC posts for reasons and purposes which are not going to be made public. In my expert opinion, results were conclusive.