How does SEM computational complexity in OpenMx scale with sample size?