I don't know about April Fools, but software pessimization as a trend has definitely been keeping Moore's law alive for a long time (via creating a demand for it).
Similarly to compiler pessimization, software pessimization operates on a higher level, and is implemented by the application programmer. The goal is to produce new software that does the exact same thing as old software, but uses more resources.
Actually I do have dynamic pessimizations steps in my compiler.
It's done when boxing raw stack values back into typed heap values. A series of ops acting on typed values might be faster with the unboxed variants, but only if all ops exist to work with unboxed values. If not, we need to pessimize back into boxed values.
It's tagged as an April Fools' joke. 1971 seems really early for that in a professional publication! Were there earlier ones?