Surprisingly (to me at least) scaling single threaded code to multiple CPUs is actually very easy: just spawn the node process N times (using cluster, naught, pm2 or similar project) and that's it.