I created pull request for Japanese Reference Manual.

Note that String#casecmp can return nil:

"\u{e4 f6 fc}".encode("ISO-8859-1").casecmp("\u{c4 d6 dc}")   #=> nil

The relevant lines in the source code (in string.c):

static VALUE
rb_str_casecmp(VALUE str1, VALUE str2)
{
    /* ... */

    enc = rb_enc_compatible(str1, str2);
    if (!enc) {
        return Qnil;
    }

    /* ... */
}

I tried to clarify this in the rdoc with r57886.