socket.io broadcast function & Redis pub/sub architecture

i would appreciate if somebody could help me with a small doubt.

Whats the difference between using socket.io broadcast function and designing the architecture with pub/sub on Redis?.

For instance, on the further example, the node.js server is listening (socket.io) CRUD requests (create) for a "key" (model 'todo'), and value 'data'. The moment it receive it, it emits again to the same user, and broadcast to all users listening on the same "channel".

socket.on('todo:create', function (data, callback) {
    var id = guid.gen()
      , todo = db.set('/todo/' + id, data)
      , json = todo._attributes;

    socket.emit('todos:create', json);
    socket.broadcast.emit('todos:create', json);
    callback(null, json);
});

But there is another way of "broadcasting" something using socket.io, and is using a pub/sub platform with Redis for key:value functions. For instance, on the further case, we are listening for a CRUD request based on a "key" (model), function (create), and value (data). But on this case, once its received, is not sent via "socket.broadcast.emit()", but published on Redis:

socket.on(key + ':create', function (data, callback) {
  var t = new ModelClass(data)
    , name = '/' + key + ':create';
  t.save(function (err) {
    pub.publish(key, JSON.stringify({key: name, data: t}));
  });
});

So on the server side, every change made on a model (and published to Redis), will be catched up (handler), and sent to the user client side (on my case, backbone.js), that will render its model according to the key:value received:

sio.on('connection', function (socket) {

   sub.on('message', function (channel, message) {
      var msg = JSON.parse(message);
      if (msg && msg.key) {
          socket.emit(msg.key, msg.data);
      }
});

So my question is very simple :-) : whats the difference of both architectures? which one is more scalable