Benchmark: timing 1000 iterations of each, for keys, foreach keys...
      each: 3.46613 wallclock secs ( 3.40 usr +  0.01 sys =  3.41 CPU) @ 292.83/s (n=1000)
  for keys: 1.98373 wallclock secs ( 1.97 usr +  0.00 sys =  1.97 CPU) @ 507.10/s (n=1000)
foreach keys: 2.02901 wallclock secs ( 1.98 usr +  0.00 sys =  1.98 CPU) @ 504.54/s (n=1000)