In any practical application, the number of GPIOs simply wouldn't be an issue. Transistors are as cheap as sand.
What would you foresee as the application for a keyboard like this? It sounds like Hall-Effect switches would work just as well and cost significantly less.
You may be right.
Thinking about it a bit more, if it were true that "in any practical application, the number of GPIOs simply wouldn't be an issue. Transistors are as cheap as sand," nobody would multiplex keyboards or LEDs at all. And it's true that it's feasible at this point to put a driver chip on every LED; the WS2812 is a common chip that you can hook up into long daisy-chains into which you shift a bunch of digital binary data to tell them how to drive one RGB LED each. And you can do the same thing with keys on a keyboard, putting one microcontroller on every key and connecting them all to a common bus or a token-ring-like bucket brigade.
But people still do multiplex lots of LED matrices and keyboard matrices.