This question was also asked on slack where I have responded.
I'm reproducing my reply (expanded slightly based on the slack conversation) here to save others the time...
Node-RED does not have any caching of the sort you are interested in. All nodes are started when the runtime starts, and are only stopped when the runtime stops or new flows are being deployed.
There is no built-in mechanism for unloading nodes that have been inactive. It is not a simple problem to identify if a node is 'inactive' - just because it hasn't sent/received a message, it doesn't mean it is inactive. For example, a node could be monitoring an external resource - the runtime has no insight into that.
It leaves memory management to the node.js runtime.
