GLib's Unicode normalization methods strongly rely on heap allocations to perform the normalization.

libunistring's normalization methods don't allocate themselves the output buffer, and thus, even stack-allocated memory can be used to perform the normalization:
http://www.gnu.org/software/libunistring/manual/libunistring.html#Normalization-of-strings

Thus, linking to libunistring to perform Unicode normalizations could really improve the performance of the parsing operations.

Also:
 * A full-text normalization instead of a word-by-word one could be done.
 * Same approach of using libunistring could be applied for casefold-ing done just before normalization.

Note: instead of libunistring, libicu is also probably a good choice:
http://bugs.icu-project.org/trac/browser/icu/trunk/source/common/unicode/unorm2.h