"There's no semantic reason my program isn't embarrassingly parallel but oh WTF are my libraries/frameworks doing? is it all a lie?" is apparently the pretty universal first experience trying to speed things up with threads. Hopefully we're working towards a future where single threaded legacies don't routinely stab the unwary in the back.
I remember in the early 2000s doubling the performance of a CPU-bound C++ program by using two threads, and my boss being shocked that I was able to do it. He had written it off as a naive idea. I was too young and inexperienced to understand why he was surprised; to me it just seemed like he didn't understand what I was doing. In retrospect, now I feel like maybe he had the right default expectation, and I got lucky that it worked out so easily.
I remember in the early 2000s doubling the performance of a CPU-bound C++ program by using two threads, and my boss being shocked that I was able to do it. He had written it off as a naive idea. I was too young and inexperienced to understand why he was surprised; to me it just seemed like he didn't understand what I was doing. In retrospect, now I feel like maybe he had the right default expectation, and I got lucky that it worked out so easily.