One of the main cool things about node based applications like houdini, is the clear visual representation of what's going on during the build of a recipe or when opening a scene made by somebody else...

Unfortunately, this is not the case with houdini scenes due to the fact that we have to keep referencing lots of object paths from/to different nodes where most of the times these nodes are even referenced inside a totally different OP network/subnet.

So, I suggest the following:
1. Allow the OP networks to be expandable (like backdrops - may be using the same UI method with different look)
2. Add a flag in the node view settings that allow hiding/showing the indirect connections between nodes references (the connection lines can be dotted with an arrow showing the direction of the reference connection, and can also be color coded make it clear if this is an object reference or attribute reference).
3. When 'tab' is pressed or the user requested a new node, check if the mouse is over an expanded OP Network, if it is, then filter the possible nodes as how already happening when requesting new node in this type of OP network.
4. Add input ports to nodes like 'object merge', 'dopimport'...etc, so the user can directly connect to them rather than referencing an object path (or maybe create equivalent nodes with nicer names since in this case, it won't be a dopimport or objectMerge, but it will be just a pass through with some parameters).