Well the current benchmarks are in (if I did it right :P )

Update: added Zaxos code.

Benchmark: timing 100000 iterations of jeffa, zaxo, original, particle
+, prakash...
     jeffa:  7 wallclock secs ( 6.36 usr +  0.00 sys =  6.36 CPU) @ 15
+725.74/s (n=100000)
     zaxo:  6 wallclock secs ( 5.15 usr +  0.00 sys =  5.15 CPU) @ 194
+28.79/s (n=100000)
  original:  4 wallclock secs ( 3.98 usr +  0.00 sys =  3.98 CPU) @ 25
+150.91/s (n=100000)
  particle:  4 wallclock secs ( 3.83 usr +  0.00 sys =  3.83 CPU) @ 26
+136.96/s (n=100000)
   prakash:  4 wallclock secs ( 3.62 usr +  0.00 sys =  3.62 CPU) @ 27
+654.87/s (n=100000)
[download]

So prakash is still in the lead :). I had to modify Zaxos code slightly to make a local copy of the array passed as its use of shift is destructive (so calling it more than once wasn't originally possible).