First time I've done a benchmark, but I think I did it right...

use Benchmark qw(timethese cmpthese);

$url = 'http://www.larryandnatalie.com/isapi.dll/c/content/f/viewprope
+rty/siteid/wb5KAM/contentclass/PICT/contentid/ZZZZZZVR/propertyname/O
+riginal/~/yamaha_58_page_11_jpg.jpg';

$r = timethese( -5, {
    'regex' => sub{$url =~ s/^(.{35})(?=.{19}).+(.{15})$/$1...$2/;},
    'substr' => sub{$url = substr($linktext, 0, 35) . "..." . substr($
+linktext, -15) if length($linktext) > 53;},
});

cmpthese $r;

__END__
Benchmark: running regex, substr for at least 5 CPU seconds...
     regex:  2 wallclock secs ( 5.07 usr + -0.07 sys =  5.00 CPU) @ 63
+3939.80/s (n=3169699)
    substr:  6 wallclock secs ( 4.69 usr +  0.34 sys =  5.03 CPU) @ 35
+32631.81/s (n=17769138)
            Rate  regex substr
regex   633940/s     --   -82%
substr 3532632/s   457%     --
[download]

It would seem as though substr is faster for doing it many times, but the regex for doing it once. Don't quote me on that though, it's quite likely that I've interpreted the results wrong... ;-)