Out-Fibbing CPython with the Plush Interpreter

A developer optimizes their custom Plush interpreter by implementing several bytecode-level improvements including instruction fusion, self-modifying code, and in-place stack operations. Starting with a recursive Fibonacci benchmark that ran slower than CPython (9.10s vs 5.70s), they achieved significant speedups through optimizations like combining push/call instructions, implementing lazy compilation with instruction patching, and creating specialized arithmetic instructions. The final result outperformed CPython at 4.57s, nearly doubling the original performance, though these microbenchmark improvements didn't translate to real-world raytracer performance gains.