use Benchmark;

my $tests = [qw(these are a few test to try
                andaverylongteststring
                acoupleoftimesover
                xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+xxxxxxxxxxxxxxxxxxxxxxxxxx
                zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz
+zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzz
zzzzzzzzzzzzzzzzzzzzzzzzzzzzz)];

timethese (100000, {
NoOpLoop =>
sub {
  for my $test (@$tests) {
    # NO Op Loop
  }
},
regex1 =>
sub {
  for my $test (@$tests) {
    $test=~s/^(.*)(.)$/\L$1\U$2/;
  }
},
substr =>
sub {
  for my $test (@$tests) {
    $test= lc(substr($test, 0, length($test)-1))
           . uc(substr($test, -1, 1));
  }
},
reverse=>
sub {
  for my $test (@$tests) {
    $test=reverse ucfirst lc reverse $test;
  }
}
});
[download]

Benchmark: timing 100000 iterations of NoOpLoop, regex1, reverse, subs
+tr...
  NoOpLoop:  2 wallclock secs ( 1.86 usr +  0.00 sys =  1.86 CPU) 
             @ 53763.44/s (n=100000)
    regex1: 82 wallclock secs (30.31 usr +  0.01 sys = 30.32 CPU) 
             @ 3298.15/s (n=100000)
   reverse:  8 wallclock secs ( 7.40 usr +  0.00 sys =  7.40 CPU) 
             @ 13513.51/s (n=100000)
    substr:  8 wallclock secs ( 8.51 usr +  0.01 sys =  8.52 CPU) 
             @ 11737.09/s (n=100000)
[download]

Hope it helps
UnderMine