Benchmark: timing 1000 iterations of glob, merlyn_1, merlyn_2...
      glob:  6 wallclock secs ( 4.55 usr +  1.55 sys =  6.10 CPU) @ 163.93/s (n=1000)
  merlyn_1:  1 wallclock secs ( 0.37 usr +  0.13 sys =  0.50 CPU) @ 2000.00/s (n=1000)
  merlyn_2:  1 wallclock secs ( 0.47 usr +  0.17 sys =  0.64 CPU) @ 1562.50/s (n=1000)
           Rate     glob merlyn_2 merlyn_1
glob      164/s       --     -90%     -92%
merlyn_2 1562/s     853%       --     -22%
merlyn_1 2000/s    1120%      28%       --