I can confirm that registers 0x17, 0x18, 0x27 and 0x28 indeed all influence hi-hat (and cymbal). Here are some notes on my reverse engineering work on this topic. It may give some insight on why this is the case.
Some historic context:
* OpenMSX offered the choice between 2 YM2413 emulation cores (originally) written by respectively Jarek Burczynski and Mitsutaka Okazaki.
* These mostly sound very good, but both are known to be not 100% accurate.
* Between 2015 and 2020 I was (slowly) working on reverse engineering the YM2413 myself.
* While doing this I documented my (partial) findings in a series of articles.
* These articles were mostly written for myself. So they might be difficult to understand for people not already deeply familiar with the YM2413.
* In 2020 we integrated the Nuke.YKY YM2413 emulation core into openMSX.
* This new core seems to be fully bit-accurate. So there was no point in continuing my own investigation.
The above link is for the article I wrote about how the YM2413 generates the different rhythm sounds (the exact waveforms). Though for all the details, you'll have to study the Nuke.YKT emulation code.
