The EXPR version's also a good bit faster since it doesn't have to do the block enter/leave setup each item.

$ perl -MBenchmark=cmpthese -le 'cmpthese( 500_000, { BLOCK => sub { g
+rep { /a/ } qw( aa bb ab cd ef ) }, EXPR => sub { grep /a/, qw( aa bb
+ ab cd ef ) }, } )'
Benchmark: 
timing 500000 iterations of
 BLOCK, EXPR
...

     BLOCK: 16 wallclock secs (16.78 usr +  0.05 sys = 16.83 CPU) @ 29
+712.16/s (n=500000)

      EXPR:  4 wallclock secs ( 3.90 usr +  0.04 sys =  3.94 CPU) @ 12
+6984.13/s (n=500000)

          Rate BLOCK  EXPR
BLOCK  29712/s    --  -77%
EXPR  126984/s  327%    --
[download]