Interesting, I'd never heard of the technique of searching all possible instruction sequences to find one with desired characteristics. It's the Infinite Monkey Theorem for code.
I ran into it a couple years ago, after hearing from other research from her. Her papers are really impressive. I would have loved that the author of the article had mentioned her.
On a side note, I wouldn't be surprised if the JIT compilers were doing some use of this, allowing some interpreted language to have better performances than compiled C on some benchmarks
EDIT : formatting and JIT addition
You'd probably freak out and worry about getting ATM card back.