Thomas Wouters suggested an obvious improvement to unicode_eq():

use memcmp().