I'd consider Bit::Vector to ensure a (supposed :) best combination of speed and platform-independence, not to mention ease of use:

use Bit::Vector;

my $size=16;
my $bits = Bit::Vector->new_Bin($size,"1101000010101111");

print "orig:\t\t",$bits->to_Bin(), "\n";
$bits->Interval_Reverse(($_-1)*8,$_*8-1) for (1..($size/8));
print "byte-rev:\t",$bits->to_Bin(), "\n";
__END__
orig:     1101000010101111
byte-rev: 0000101111110101
[download]