Sorry if I missed it, but does Otter do anything special to limit GC-impact when caching lots of references? That's the main selling point for bigcache and freecache.
The number of items isn't the best metric alone, making it harder to demonstrate in a benchmark. I've reaped benefits from bigcache with ~600k-900k items — but there's a lot of eviction, each item is a pointer and can have a bunch of nested references itself. (protobufs)