I did something similar to reduce the memory overhead of our Rust backend at a recent job; I noticed that when dealing with certain requests, we were using quit

I did something similar to reduce the memory overhead of our Rust backend at a recent job; I noticed that when dealing with certain requests, we were using quite a lot of memory on duplicates of the same strings when processing things, so I made something I called an "alias cache" type that could be used to obtain an alias of an existing string if needed, and then introduced those to several places in the codebase where it was most helpful. This strategy also allowed some flexibility on how long strings were "cached" for, which let me tweak which layers of processing should share a cache versus creating their own (and letting the old one drop). Our use case definitely didn't have room for interning to use 2000x less memory, but I think in some of the pathological cases it reduced the max memory usage to around 50% of what it was before without significantly affecting runtime, which I was pretty pleased with!