The Node Is Nonsense: Better ways to measure progress than Moore's law [pdf]

2000x is believable, but that doesn't mean the latest algorithm will run on Apple II. Algorithmic speedup is often hardware relative. For example better cache locality is less important in older hardwares.

On the actual back-in-the-day Apple II at time of release, you probably did not get a configuration that maxed out RAM and storage because it just cost too darn much. But as you got into the 80's, prices came down, and the Apple could expand to fit with more memory and multiple disks, without architectural changes.

As such, a lot of the early microcomputer software assumed RAM and storage starvation and the algorithms had to be slow and simple to make the most of memory, but Apple developers could push the machine hard when ignoring the market forces that demanded downsizing. It was a solid workstation platform. When the 16-bit micros came around the baseline memory configuration expanded remarkably, so completely different algorithms became feasible for all software, which made for a substantial generation gap.

By the 90's, the "free lunch" scenario was in full swing and caching became normalized, and so everything has been about faster runtimes, but at systemic scale, with layers of cache, it's often hard to pinpoint the true bottleneck.