When presented with an input file which has a 2-byte ToUnicode CMap pdfwrite only emits a 1-byte ToUnicode CMap, and only containing those character codes which are valid for one byte. The problem is that we don't treat Type 0 fonts the same as CIDFonts (for ToUnicode), instead we handle them as type 1, which can only have 256 entries
Fixed in revision 16696ae, patch here: http://ghostscript.com/pipermail/gs-cvs/2011-June/013005.html