I'd bet that during those "episodes" the load average spikes; you could just 
do a looped "uptime" and watch the figures on a console.