+            // Yield here to avoid pinning the CPU. Give other threads some CPU
+            // time to avoid a spiky latency distribution (BF-27452). Even if
+            // this client can run continuously and receive another command
+            // without blocking, we yield anyway. We WANT context switching, and
+            // we're trying deliberately to make it happen, to reduce long tail
+            // latency.
+            _yieldPointReached();
             _receiveRequest();