HTTP/2 initially seemed promising for reducing latency, but caused intermittent slowdowns for the client's critical assets. Testing revealed HTTP/2 increased median page latency by 20-30% compared to HTTP/1.1. While HTTP/2 performs well in some scenarios, it was not suitable for delivering critical assets globally due to variance issues. Over time the performance of HTTP/2 has improved, but introducing artificial latency still results in a 2.3x slowdown, so HTTP/1.1 remains a better solution for this client's critical path resources.