According to @_avichawla, LMCache’s multiprocess caching and CacheBlend deliver 14x faster TTFT and 4x faster decode on Qwen3-235B, cutting costs by up to 90%.