When storing Unicode data in varbinary columns the binary stream must begin with the appropriate marker (0xFFFE), otherwise the iFilter will not be able to properly parse the content when building a full-text index. I've learned this the hard way. Well, it didn't actually hurt, but it took a day or two to get cleared up. The interesting bit is the fact that when Unicode data is cast to XML prior to storing it in a varbinary(max) column, the stream is properly marked automatically. This made me think the whole thing was a bug, now it proves to be "by design". One new lesson learned: Need to construct unicode file properly.