For some software, higher performance is more valuable than fewer lines of code. For example, if you re-wrote the Linux kernel in Lisp, it might have many fewer lines of code than the current C version, but would deliver less value because it would run much slower and might stop for garbage collection in the middle of an interrupt handler. Concerns about real-time performance and memory-usage would also apply to lots of embedded software, like the code that controls the engine in your car.

I'm not sure that I'd call something a "fundamental theorem of software development" if there are large classes of software that it doesn't apply to.