The fact that it split the chapters in seemingly arbitrary places indicates that the xpath didn't work - when that happens the book is split ever 270K.
The example text you show should work fine with the xpath above, but by your wording I get the sense that the example text is not from the book which was exhibiting problems. My guess is the book with problems doesn't use h1/h2.
You read the manual already, but this tutorial might help:
https://www.mobileread.com/forums/sho...d.php?t=129364