Im sorry to be jumping in this so late...
This is an interesting problem, but I think that it is important to accept that the current system is a hack-job as well. The current system is the equivalent to "gotten from page 57 of the book downloaded from
www.something.com version 52.01b on sept 15 2005, viewed on a reb 1150, at medium font size, using default fonts." The exact printing, and page number have to be referenced, as well as enough information to distinguish the work. All this mess to get what level of granularity? You only get within a few paragraphs! So the first question in looking forward is what level of granularity do you want to have, and what should it be based on. Looking back the only real option is to have a page by page reference (paper or electronic) of every thing that has ever been printed... It is a broken system!
As far as reference, I think that a system of smaller granularity chunks until the desired resolution is gained would be idea. Choices like what sorts of content are "valid" at each level would be important, but in normal works chapter, paragraph, and then character range would be possible. Section, sentence, word would be similarly reasonable. The most important part of this would be having a standard. Having a standard edition to pull from is similarly important.
If a standard was adopted, the first big sign that it was being used would be that academic, printed journals would adopt the numbering in their publications. I think this would be just amazing... as the current system was accepted based on its retrospective convenience, and little else.
If such a system was adopted, I think the next question would be, how would you update all of the archaic references in previously printed works?
Life and times of Bilbo Baggins, page 54
might become
Life and times of Bilbo Baggins, v1.00: chapter 2: paragraph:16-19
thats not all bad is it?