Inference Performance Overhead

Latency, throughput, and cost penalties of encrypted or otherwise protected inference relative to plaintext baselines.