We forgot to free the TCB allocated buffer on thread destruction, causing
a leak of around 1000 bytes per thread creation. We still have another
leak (exactly one page per thread object creation) that I'm trying to find :(