They have a template set (we have a port of that in our designer) called self-tracking entities (STE's). They serialize the changes with them. All other types are poco's and you work with derived classes created at runtime.
That they don't do this is indeed a downside, if you want to use the entities on the client side, alter changes there and send them over to the service.
So as you don't have that feature, one often opts for a different model on the client, or generate service layers through e.g. OData (which effectively creates a new model on the client). It's more work indeed. Especially as they also lack a disconnected unit of work, and don't have features like our datascopes which make using entities on the client easy and you can send the unit of work over the wire to the service with a few lines of code.