Global replace across multiple files has an unusual behavior if you're searching for some characters and replacing it with some other characters.

Test case -- make two files.  They are similar save for the fact that the latter has one less character.  Note that the lines end with the "registered" symbol, or the "trademark" symbol.

First file:
FOO BAR®
FOO BAR™
FOO BAR®
FOO BAR™

Second file:
FOO BAR
FOO BAR™
FOO BAR®
FOO BAR™

Do a Search and Replace for TM, replacing it with ™.

The second replacement on the second file will look like this:

FOO BAR
FOO BAR™FOO BAR®
FOO BA™

I'm suspecting that a char offset is being used on multibyte strings, or something like that.