Hi @DiaAzul,
wow thank you very much for your lengthy and detailed answer!
In fact I did some similar marshalling and unmarshalling already in the past (Load a table once and use it at several different points in the workflow).
However, as you said this is not working for all data types and might be error prone. A standard implementation of creating a pointer to the table as a variable and a subsequent dereferenciation would be way more robust and, in my ideal world, wouldn’t be too complicated to implement in the background of KNIME 
Best,
Dominic