I think the document could be a little more explicit about the sequence constraints in the protocol, e.g. where blocking
 is required and where pipelining of data is allowed.

For example, it seems that it should be OK for the client to send a GET message with the trailing certificate request 
without waiting for the response message, and on success the server can send the response message, cert chain, and 
second response message. This reduces the round-trip delays and seems a useful thing to make explicit in the 
specification as either allowed (or disallowed and the reason why).

Also, if pipelining like this is allowed, it would be useful to give recommendations for handling of failures, e.g. is 
there any reason the client should destroy the private key used in a failed request, or is it a safe (security-wise) 
optimization to reuse that key in a subsequent request?

--Karl Czajkowski