The second argument to g_once_init_leave() must be non-zero.
In environments which #define the clockid to be used (CLOCK_MONOTONIC or CLOCK_REALTIME) as 0, the current code in g_get_monotonic_time() triggers an assertion in g_once_init_leave() and apparently can not function as expected.

A possible fix might be like

--- glib/gmain.c.ori    2012-01-25 21:27:35.000000000 +0100
+++ glib/gmain.c        2012-01-25 21:38:43.000000000 +0100
@@ -2043,22 +2043,21 @@
   /* librt clock_gettime() is our first choice */
   {
 #ifdef HAVE_MONOTONIC_CLOCK
-    static volatile gsize clockid = 0;
+    static volatile gsize clockid = CLOCK_REALTIME;
+    static volatile gsize clockid_guard = 0;
 #else
     static clockid_t clockid = CLOCK_REALTIME;
 #endif
     struct timespec ts;
 
 #ifdef HAVE_MONOTONIC_CLOCK
-    if (g_once_init_enter (&clockid))
+    if (g_once_init_enter (&clockid_guard))
       {
-       clockid_t best_clockid;
-
        if (sysconf (_SC_MONOTONIC_CLOCK) >= 0)
-         best_clockid = CLOCK_MONOTONIC;
+         clockid = CLOCK_MONOTONIC;
        else
-         best_clockid = CLOCK_REALTIME;
-       g_once_init_leave (&clockid, (gsize)best_clockid);
+         clockid = CLOCK_REALTIME;
+       g_once_init_leave (&clockid_guard, (gsize)1);
       }
 #endif