Yes, the lack of maintenance of gumbo-parser was another reason I decided to include a private copy. You are welcome to make your PR against html5-parser's copy of gumbo-parser, and I will review it.
There's quite a few changes on the original upstream since then. gumbo_malloc -> gumbo_parser_allocate, free_node -> destroy_node, passing around the parser as an argument, &c.
I have no objectsion to merging in changes from gumbo-parser that are not in sigil-gumbo over time, but it will need to be done gradually, as there is only so much time I can devote to html5-parser now that it meets the needs of calibre.
I have some doubts that html5-parser would be 30x faster than html5lib without the performance work in 0.9.4; Gumbo up through 0.9.3 was really slow for a C library.