Thanks for all the great answers, especially those with code samples and benchmarks! It looks like the syntax and space efficiency winner is Set::Light, though not by much (it uses a C module to point all the set members to a single object).

Remember I'm in mod_perl, so construction/deconstruction time is less important than memory footprint and lookup time.

#! /usr/bin/perl -w
use strict;
use Devel::Size qw(total_size);
use Set::Light;

my $lookup = "shave";

my %hash1 = (
    shave  => '',
    the    => '',
    very   => '',
    modern => '',
    way    => '',
);

my %hash2 = (
    shave  => 1,
    the    => 1,
    very   => 1,
    modern => 1,
    way    => 1,
);

my %hash3 = (
    shave  => undef,
    the    => undef,
    very   => undef,
    modern => undef,
    way    => undef,
);
print "Hash3 " . ((exists $hash3{$lookup})?"contains":"does not contai
+n") . " $lookup\n";

my $foo = undef;
my %hash4 = (
    shave  => \$foo,
    the    => \$foo,
    very   => \$foo,
    modern => \$foo,
    way    => \$foo,
);

my $set = Set::Light->new( qw/shave the very modern way/ );
print "Set " . (($set->has($lookup))?"contains":"does not contain") . 
+" $lookup\n";

print "Size 1: " . total_size(\%hash1) . "\n";
print "Size 2: " . total_size(\%hash2) . "\n";
print "Size 3: " . total_size(\%hash3) . "\n";
print "Size 4: " . total_size(\%hash4) . "\n";
print "Size S: " . total_size(\$set) . "\n";
__END__
[download]

Hash3 contains shave
Set contains shave
Size 1: 363
Size 2: 303
Size 3: 283
Size 4: 315
Size S: 251