Here are some basic user comments.
When using fonttools (namely
pyftsubset), OpenType features are preserved with the embedded font. See the joint screenshot of a pdf file made by Prince from an ePub3. You can spot the words "a
ffaire" (twice) and "
Quelles" (the Q with a long tail). As Doitsu said, it does not mean that they will be displayed with every reader, but, up to now I did not spot any two-letters "hole".
My use case revolves around this. I willingly recognize that I do not care if every feature of my ePubs cannot be downgraded to Mobi level but that does not imply that it cannot be done, just that I did not try it (though I own a Kindle, I read only ePubs and PDFs using Koreader).
For your information, here is a link which further details the
use case for pyftsubset. It tells that ligatures are preserved.
If you wish to check by yourself -which is always a good idea-, here are the fonts I've used for these trials (resources.zip). The regular, italic, bold and bolditalic fonts are forked from the 5.3 version of Linux Libertine (and of course share the same license). The joint SC variant, that I use for convenience purposes, comes from a subset I obtained using FontSquirrel some years ago. You can forget it.