Sure I completely get that, but what the core issue I was hitting on is that you end up trading an expensive resource (computing power) for a very cheap resource (disk space). In cases where disk space requirements become enormous this can be a necessary trade, but here that's not really an issue. I enjoy not strictly necessary optimization as much as anybody, but unless I'm missing something this optimization seems that it would ultimately result in a less performant system with very little to show for it so it wouldn't be particularly fair to call it an optimization.

Ah yes, but this way you get to write compression algorithms and do performance optimization. The journey is more important than the destination.

Edit: Also (in Lichess' case) you're not trading CPU usage for disk space, you're trading CPU usage for disk space and network traffic, which you're probably paying much more for.