It's high time that programming under resource constraints was rediscovered and valued. It's been so easy to throw hardware (or someone else's hardware, ie cloud) at problems that efficiency and cost per compute output has been forgotten. Maybe a few years of serious memory shortage will focus minds. Perhaps this is one of those things that needs to be recast in terms a CFO can understand before it gets attention: "you're paying XXX per compute node because your team writes flabby software. You could save $$$ if they fit into Y instead"