I think the below is functionally the same as yours:

#!perl -l
my %weights=(a=>1,b=>2,c=>3);

my $max=0;
my @values=map { my $range=[ $max, $max+$weights{$_}, $_ ]; 
                 $max+=$weights{$_}; $range } keys %weights;

foreach (1..100) {
    my $r=rand($max);
    $_->[0]<=$r and $r<$_->[1] and print $_->[-1]
        foreach @values;
}
[download]

You could binsearch the list instead of linsearching it for a bit of speed up.