TL;DR When talking about caching, is it even worth considering try and br as memory efficient as possible?

Context:
I recently chatted with a developer who wanted to improve a frameworks memory usage. It's a framework creating discord bots, providing hooks to events such as message creation. He compared it too 2 other frameworks, where is ranked last with 240mb memory usage for a bot with around 10.5k users iirc. The best framework memory wise used around 120mb, all running on the same amount of users.

So he set out to reduce the memory consumption of that framework. He alone reduced the memory usage by quite some bit. Then he wanted to try out ttl for the cache or rather cache with expirations times, adding no overhead, besides checking every interval of there are so few records that should be deleted. (Somebody in the chat called that sort of cache a meme. Would be happy , if you coukd also explain why that is so😅).
Afterwards the memory usage droped down to 100mb after a Around 3-5 minutes.

The maintainer of the package won't merge his changes, because sone of them really introduce some stuff that might be troublesome later on, such as modifying the default argument for processes, something along these lines. Haven't looked at these changes.

So I'm asking myself whether it's worth saving that much memory. Because at the end of the day, it's cache. Imo cache can be as big as it wants to be, but should stay within borders and of course return memory of needed. Otherwise there should be no problem.
But maybe I just need other people point of view to consider. The other devs reasoning was simple because "it shouldn't consume that much memory", which doesn't really help, so I'm seeking you guys out😁