This node can be used for performance optimization. If you have a complex or heavy calculation with result that's read by multiple other nodes, you can place this node at the end. This will make sure that the input is evaluated only once (per evaluation context) and the result reused for the other nodes.
In most cases you don't need it and it doesn't change the functionality in any way. At some point in the future this is going to be handled automatically.