I think your benchmark is flawed. I don't think your append code is doing what you think it's doing. Witness:

    my $buffer .= "x" x 1024 for 0 .. 5;
    print length $buffer;
[download]

$buffer is uninitialized by the time it reached the print. Moving the declaration away from the assignment fixes this.

Furthermore, both fragments don't produce the same result. The 'push' variant keeps everything in an array - it ought to have final join, IMO.

Here's a somewhat different benchmark. It also shows the append method to be faster, and much faster than I expected it to be.

#!/usr/bin/perl

use strict;
use warnings;

use Benchmark qw /cmpthese/;

our $str;
              
our ($a, $b);

for our $t (5, 10, 100) {
    for my $n (1, 8, 64) {
        $str = "x" x ($n * 1024);
        cmpthese -3, {
            "append${n}k_$t" => '$::a = ""; $::a .= $::str for 1 .. $:
+:t;',
            "push${n}k_$t"   => 'my @s;  push @s => $::str for 1 .. $:
+:t;
                                 $::b = join "" => @s',
        };
        die "Benchmark is wrong\n" unless length ($a) == $t * length $
+str &&
                                          length ($b) == $t * length $
+str;
    }
}

__END__
Benchmark: running append1k_5, push1k_5 for at least 3 CPU seconds...
               Rate   push1k_5 append1k_5
push1k_5    91330/s         --       -58%
append1k_5 219880/s       141%         --

Benchmark: running append8k_5, push8k_5 for at least 3 CPU seconds...
              Rate   push8k_5 append8k_5
push8k_5   33834/s         --       -50%
append8k_5 67846/s       101%         --
 
Benchmark: running append64k_5, push64k_5 for at least 3 CPU seconds..
+.
              Rate   push64k_5 append64k_5
push64k_5    460/s          --        -92%
append64k_5 5705/s       1139%          --
 
Benchmark: running append1k_10, push1k_10 for at least 3 CPU seconds..
+.
                Rate   push1k_10 append1k_10
push1k_10    53627/s          --        -60%
append1k_10 135692/s        153%          --
 
Benchmark: running append8k_10, push8k_10 for at least 3 CPU seconds..
+.
               Rate   push8k_10 append8k_10
push8k_10   16317/s          --        -67%
append8k_10 48795/s        199%          --

Benchmark: running append64k_10, push64k_10 for at least 3 CPU seconds
+...
               Rate   push64k_10 append64k_10
push64k_10    221/s           --         -85%
append64k_10 1510/s         583%           --
 
Benchmark: running append1k_100, push1k_100 for at least 3 CPU seconds
+...
                Rate   push1k_100 append1k_100
push1k_100    5730/s           --         -60%
append1k_100 14175/s         147%           --
 
Benchmark: running append8k_100, push8k_100 for at least 3 CPU seconds
+...
               Rate   push8k_100 append8k_100
push8k_100    185/s           --         -84%
append8k_100 1181/s         537%           --
 
Benchmark: running append64k_100, push64k_100 for at least 3 CPU secon
+ds...
                Rate   push64k_100 append64k_100
push64k_100   22.5/s            --          -76%
append64k_100 94.7/s          320%            --
[download]

Abigail