What sort of problems would you need to be doing in Python that aren't already done by libraries like NumPy which would require that sort of mental model?
The only thing that comes to mind would be memory leaks in long running applications caused by your data structures having some link to a link to a link that prevents temporary data from being garbagecollected.