First off - has anyone done this?

Yup. One of the motivations behind Test::Class was solving exactly this sort of problem. Getting test running is pretty simple.

# create a base class that contains your generic tests
{   package BaseClass::Test;
    use base qw( Test::Class );
    use Test::More;
    
    sub class_under_test {
        my $self = shift;
        my $class = ref $self;
        $class =~ s/::Test$//;
        return $class;
    }
    
    my $o;
    
    sub create : Test( setup => 1 ) {
        my $self = shift;
        my $class = $self->class_under_test;
        $o = $class->new;
        isa_ok $o, $class;
    }
    
    sub answer : Test {
        is $o->answer, 42;
    }   
    
    # because we don't want our abstract test class to run
    __PACKAGE__->SKIP_CLASS( 1 ); 
}

my @classes_to_test = qw( Foo Bar Ni );

# make a bunch of subclasses of our base test class
eval "package ${_}::Test; use base 'BaseClass::Test'"
    foreach @classes_to_test;

# run everything
Test::Class->runtests;

__END__
# will output something like

1..6
ok 1 - The object isa Bar
ok 2 - answer
ok 3 - The object isa Ni
not ok 4 - answer
#   Failed test 'answer'
#   in foo.pl at line 44.
#          got: 'infinity'
#     expected: '42'
ok 5 - The object isa Foo
ok 6 - answer
# Looks like you failed 1 test of 6.
[download]

Where it currently falls down is the test reporting. It's currently a little tricky to see which particular subclass failed a test. You either have to add it in to the test description explicitly, or set TEST_VERBOSE - which is a little too verbose. Making this easier is on the to do list.