Preventing resource starvation on HTTP GW


I am running a bunch of Kubo HTTP gateways.
They are 32vCPU , 128GB RAM, 20TB volume and zero PINS - just caching.

Server profile.
AcceleratedDHTClient is disabled.

Things run smoothly until I get hit with a spike of local CID misses.
RAM usage ballons to 100GB+ and 32vCPUs start spinning at 100%.

wantlist is no greater than 500.
swarm peers jumps to 100k +

IPFS soon OOMs and everything works for the next 5 minutes until the RAM/CPU goes into overdrive and next OOM hits.

Any config suggestions that could prevent me from resource starvation in case of a flood of requests for missing CIDs ?