g_cluster was attempting to deallocate a gmx_rmpbc_t object when it
might not have been initialized. Since that object is only initialized
under a simple combination of conditions, it makes most sense to clean
up only under those conditions (as well as fix the real reason for the
segfault).