Say you do most of your testing on a particular platform, browser, or mobile device. You can set up a virtual grid node that has multiple instances of that platform, browser, and device. But, after 5-10 instances, the virtual machine may run out of memory.

So, you could clone the VM, create a second identical node on the grid, and let the Selenium hub load balance the tests that get started and run on that particular platform.

The Selenium hub keeps track of which nodes are idle, and once a node has the max number of instances running on it, the hub will either add a waiting test suite to a queue or distribute it to a node with the same platform and browser/mobile device if it is found. The user doesn't have to direct it to the other node, the Selenium hub will manage the traffic flow.

But if the user varies something on the cloned node, such as the browser version, then they can actually direct traffic to each of the nodes by passing in a different browser version, say, as a parameter to the test suite!