I did some more checks and downloaded the source code of fbreader. The program uses indeed the TeX algorithm, which under normal circumstances works great. So the problem is not the hyphenation algorithm but the
detection of the language. Depending on the language the hyphenation will differ so if the wrong language is picked you will get a badly formated output. You can try it in the
online hyphenator (e.g. "clear").
In the latest OpenInkpot snapshot it's possible to correct the language. Once I did this, everything started to work perfectly!
No need to change anything.
