This issue may extend beyond just unicode.upper() and unicode.lower(), but it's very clear with these two methods, at least.

For example, consider DESERET SMALL LETTER EW.  On a UTF-16 build, calling upper on a string containing this doesn't change it to the capital variation (DESERET CAPITAL LETTER EW):

>>> u'\N{DESERET SMALL LETTER EW}'.upper() == u'\N{DESERET SMALL LETTER EW}'
True

It can also be seen that this isn't even recognized as lower case:

>>> u'\N{DESERET SMALL LETTER EW}'.islower()
False

With a UTF-32 build, however, the expected behavior (ie, the behavior one would get for a code point in the BMP with small and capital variations) is provided.