Could this be the reason I see a lot of still reachable memory when running the
panel under valgrind? panel_background_monitor_* stuff is in the trace at least.

Example:

==6647== 2359296 bytes in 1 blocks are still reachable in loss record 8538 of 8538
==6647==    at 0x1B909222: malloc (vg_replace_malloc.c:130)
==6647==    by 0x1C30A109: g_try_malloc (gmem.c:194)
==6647==    by 0x1C10B809: gdk_pixbuf_new (gdk-pixbuf.c:293)
==6647==    by 0x1C08F781: gdk_pixbuf_get_from_drawable (gdkpixbuf-drawable.c:1278)
==6647==    by 0x807F918: panel_background_monitor_get_region
(panel-background-monitor.c:356)
==6647==    by 0x807E3CD: get_desktop_pixbuf (panel-background.c:194)
==6647==    by 0x807E56F: get_composited_pixbuf (panel-background.c:245)
==6647==    by 0x807E601: panel_background_composite (panel-background.c:306)
==6647==    by 0x807E6DE: panel_background_transform (panel-background.c:478)
==6647==    by 0x8065066: panel_widget_set_background_region (panel-widget.c:1295)
==6647==    by 0x8065979: toplevel_configure_event (panel-widget.c:1584)
==6647==    by 0x1BDCD758: _gtk_marshal_BOOLEAN__BOXED (gtkmarshalers.c:83)
==6647==    by 0x1C2912FB: g_closure_invoke (gclosure.c:437)
==6647==    by 0x1C2A0A7E: signal_emit_unlocked_R (gsignal.c:2488)
==6647==    by 0x1C2A1E2B: g_signal_emit_valist (gsignal.c:2257)
==6647==    by 0x1C2A23E2: g_signal_emit (gsignal.c:2291)
==6647==    by 0x1BEAF04D: gtk_widget_event_internal (gtkwidget.c:3703)
==6647==    by 0x1BDCAF3F: gtk_main_do_event (gtkmain.c:1372)
==6647==    by 0x1C0AFD9C: gdk_event_dispatch (gdkevents-x11.c:2262)

Fixed in HEAD. I'm not sure if this is related, Kjartan. If it's still
happening, then another bug would be better :-)

Thanks