Is device polling enabled? The load average is a global number of runnable-but-waiting processes, but you should divide by the number of cores to get something more intuitively meaningful