Given the model of memory we are discussing (a global per-process LRU cache), that’s exactly what I was discussing using. Unless there’s another way to handle such global caches.