Data segment memory corruption (non-observable under debugger)

UTF-8 conversion of help files fails if the help page is fetched from the help ZIP. UTF-8 characters not in ASCII then get converted to ?. Does not occur if the help page is read from a .html file.

Somehow the UTF conversion table seems to get corrupted (data segment memory corruption).

Bonus challenge: same binary runs perfectly fine when invoked under the Watcom debugger. Will be a pain in the ass to debug. I'll try my luck running this under lDebug.

This is for example for the new french help files, as the 1.0.9 files are UTF-8. 1.0.8 files are NOT UTF-8 encoded.