Benchmark: timing 100 iterations of no ref, ref...
 no ref: 33 wallclock secs (28.19 usr +  0.07 sys = 28.26 CPU) @  3.54/s (n=100)
    ref: 10 wallclock secs ( 9.21 usr +  0.34 sys =  9.55 CPU) @ 10.47/s (n=100)