Algorithms fail and tend to get stuck in circular loops doing the last few corrections on pages --- a particularly intractable problem for H&J is the detection and elimination of multiple word stacks.
ePubs could and should be better, but until we get true A.I. (or perhaps even Self Aware) machines which are able to determine which form of a word is being used so as to determine the correct hyphenation point), they'll never be perfect and a person will always be able to do better.
Some .pdfs which should be well-done:
-
http://www.mikebrotherton.com/novels/stardragon.pdf (I did that a long while ago)
-
http://people.umass.edu/klement/russell-imp.html
Hmm, I've just realized that the loss of my Verizon hosting means that my copy of _The Book of Tea_ is now off-line....