This isn't strictly an issue for the hypertext tools that we know now-- html, Tinderbox, etc. It's really more a matter of markup and structure. Even in a linear printed document, if you have headings for chapters, topics or subtopics, or even page numbers, then an annotator writing even in a separate work can point more accurately to the location in your work that they are discussing. It's better to have a meta-structure (something which reveals the structure of your thought) like chapter headings than an arbitrary markup like page numbers, but either way you have anchors for annotation.
That structure is almost built into hypertext. Unless the author is working hard not to make their lexia correspond to discrete thoughts, then a hypertext has some sort of anchor structure to hook into for annotation. The granularity is smaller than "the entire work" if you break the work down into lexia or if you use anchors or headings to give the work some sort of structure. I say almost because it's easy to build a web page that doesn't have anchors, or headings in proper HTML. Many works on the web divide works depending on the length of the text, or how many advertisements they need to fit into the reading. This isn't entirely the fault of the authors-- most web tools make it far too easy to produce pretty documents with no meta-structure. They focus on presentation to the extent of making it possible to make a document look marked up ( for headings) without being marked up in a hypertextual way ( or even ).
Still, if the tools are built well, and are used moderately correctly, then the lexia, pages, sections, or chapters will not only correspond to thoughts, but those thoughts will be more easily referenced... and visible.