MySQL AB has confirmed one or more bug(s) when CHECKSUMs are generated on UCS2 data.
The bug is described like:
1) Different results between InnoDB and MyISAM tables when using md5() and ucs2, with GROUP BY.
2) md5() function outputting data in wrong charset (sometimes ???? -characters are returned to the client)
The bug report is now available in the public bugs.mysql.com system here.
It affects SQLyog DATA sync sometimes when working with UCS2 data. And sometimes in a weird and unpredictable way (changing a few data or omitting a single column with the <columns> options may make the difference in some situations)
MySQL AB has proposed the 'workaround' to unhex(hex(data)) in the SJA code. We are testing this. If it works as expected and if it does not affect performance too much it will be implemented, so that SQLyog DATA SYNC will work with UCS2 data and MySQL versions where this bug has not been fixed.