I'm a pragmatist, and see nothing wrong with supporting $obj->new(). In fact, perlobj even mentions it. If your class takes args that are in addition to what the parent class uses in ->new(), then you need a bit more code.

  # Create a new object
  sub new
  {
      my $thing = shift;
      my $class = ref($thing) || $thing;

      my $self = {};
      bless($self, $class)

      if (! $self->_init($thing, @_)) {
          # Failed to initialize
          # Throw some sort of error, or
          return;   # Returns 'undef'
      }

      return ($self);
  }


  # Initialize a new object
  sub _init
  {
      my $self  = shift;
      my $thing = shift;

      # Separate '@_' into args for parent class and args for this sub
+class
      my @parent_args = ...;
      my @my_args     = ...;

      # Perform parent class initialization
      if (! $self->SUPER::_init($thing, @parent_args)) {
          # Parent class initialization failed
          return (0);
      }

      # Perform subclass initialization
      # Making use of '@my_args', if any
      if (ref($thing)) {
          # $thing->new( ... ) was called
          # Make use of '@my_args', if any
          # And make use of object's data, if applicable

      } else {
          # CLASS->new( ... ) was called
          # Make use of '@my_args', if any
      }

      return (1);
  }
[download]

Note that $thing is passed to _init() in addition to the args (cf. jamesduncan's code above). When the user calls $obj->new(), this allows the initialization method to make use of data contained in the calling object, if that is applicable to your code.