I've got a couple of possibilities here for speeding up your code. Best performance thus far: a capturing regular expression:

#!/usr/bin/perl
use strict;
use warnings;
use Benchmark qw(cmpthese :hireswallclock);

my $coords = 'lon,lat,alt ' x 20000;
cmpthese(100, {
                OP => sub {
                                my @list =  split(/[, ]/,$coords);
                                my $size = @list;
                                my @poly = ();
                                
                                for (my $i=0; $i<=($size-3); $i+=3)
                                {
                                    my $lon = $list[$i+0];
                                    my $lat = $list[$i+1];
                                    my $point = [$lon,$lat];
                                    push (@poly, $point);
                                }
                            },
                double_split => sub {
                                my @list =  split(' ',$coords);
                                my @poly;
                                $poly[@list-1] = 0;
                                for (@list) {
                                    push (@poly, [ (split /,/, $_)[0, 
+1] ]);
                                }
                            },
                while_shift => sub {
                                my @list =  split(/[, ]/,$coords);
                                my @poly = ();
                                while (@list) {
                                    push @poly, [shift @list, shift @l
+ist];
                                    shift @list;
                                }
                            },
                regex => sub {
                                my @poly = ();
                                while ($coords =~ /([^,]+),([^,]+),[^\
+s]+ /g) {
                                    push @poly, [$1,$2];
                                }
                            },
                chromatic => sub { my @poly = map { [ (split /,/, $_)[
+0, 1] ] } split / /, $coords;},
                jwkrahn => sub { my @poly = map [ ( split /,/ )[ 0, 1 
+] ], split ' ', $coords;},
                ike1 => sub { my @poly = map [ (split /,/, $_)[0, 1] ]
+, split ' ', $coords;},
                ike2 => sub { my @poly = map [ /([^,]*),([^,]*)/ ], sp
+lit ' ', $coords; },
                });
__END__
               Rate   OP while_shift double_split ike1 jwkrahn chromat
+ic ike2 regex
OP           6.14/s   --        -17%         -49% -50%    -51%      -5
+2% -54%  -59%
while_shift  7.42/s  21%          --         -38% -40%    -40%      -4
+2% -45%  -51%
double_split 11.9/s  95%         61%           --  -3%     -4%       -
+7% -11%  -21%
ike1         12.3/s 101%         66%           3%   --     -1%       -
+4%  -8%  -18%
jwkrahn      12.4/s 103%         67%           4%   1%      --       -
+3%  -8%  -18%
chromatic    12.9/s 109%         73%           8%   4%      3%        
+--  -4%  -15%
ike2         13.4/s 119%         81%          13%   9%      8%        
+5%   --  -11%
regex        15.1/s 147%        104%          27%  23%     22%       1
+8%  13%    --
[download]

Update: Following ikegami's post below, I reran with more tests, closing some apps I had open and changing the spec to match the OP (originally ran for x 1000, whereas spec says "tens of thousands") and got the same ordering.

This is perl, v5.8.9 built for MSWin32-x86-multi-thread
(with 9 registered patches, see perl -V for more detail)
[download]