This is somewhere between a bug and an enhancement. When moving files with Unicode names between different OSes, it would make life a lot easier if there was an option to normalize filenames to a particular unicode form on the destination. For example, rsyncing from OS X to Linux names the files using the OS X standard, Normalization form D. Linux prefers Normalization form C, which means sharing this folder via Samba results in a number of folders that can't be accessed. Likewise when running rsync from OS X to a Windows box running cygwin. I think the best solution to this would be something like "--destination-charset=UTF-8-NFC", to rename the files from the source charset to the destionation charset.
See bug #2790 comment #3 for a link to a patch that provides charset conversion. I'd like feedback on how well this works. *** This bug has been marked as a duplicate of 2790 ***