Inlining a common case gives another 1% improvement in load CPU. Not

much, but I'll take it. :)

src/ZEO/zrpc/smac.py

@@ -266,7 +266,15 @@ class SizedMessageAsyncConnection(asyncore.dispatcher):
             while (size <= SEND_SIZE) and messages:
                 message = messages[0]
                 if message.__class__ is str:
-                    size += self.__message_output(messages.pop(0), output)
+                    if self.__hmac_send:
+                        size += self.__message_output(messages.pop(0), output)
+                    else:
+                        # inline common case
+                        lmessage = len(message)
+                        output.append(struct.pack(">I", lmessage))
+                        output.append(messages.pop())
+                        size += 4 + lmessage
+
                 elif message is _close_marker:
                     del messages[:]
                     del output[:]