This test is locale-depending. string.letters is the collection of 8-bit characters for which islower() or isupper() is true. On most Unixes the locale is either UTF-8 or ASCII and string.letters contains only ASCII letters. On most Windows it contains characters from extended ASCII, and the length of string.letters can be larger than 75 (minimal non-replacement index used in tests).

There is two simple way to fix this issue.
1) increase test indices to be larger than 255.
2) use string.ascii_letters instead of string.letters.

Here is the patch which uses the second approach (because string.letters is used in other tests).