Which I'm confused about - wouldn't decreasing the cache TTL increase compute demand?