In this case, I don't think it does. The model is equivalent to a model that has arrows from the y1 and y2 to the two first order latent variables.
Here's how I think about it: If that latent variable wasn't there, you'd have four parameters - from the two Y vars to the two latents. When you add a latent variable you are specifying some constraints (usually) on what those four parameters might be. But when you add this latent, you still have four parameters - so you are not adding any constraints - the two models are equivalent.
(Also, your Y variables are exogenous - it would be clearer if these were called X, and your X variables are endogenous. It would be clearer if these were called Y).
jeremy