It appears that this issue is grizzly specific. It's (accidentally) fixed in havana already.
In havana, we use a single queue for all replies. We're not creating an exchange/queue for every reply. This behavior was actually introduced in grizzly, but is off by default. If you set amqp_rpc_single_reply_queue=True in your config file, you should not be seeing this problem.
It appears that this issue is grizzly specific. It's (accidentally) fixed in havana already.
In havana, we use a single queue for all replies. We're not creating an exchange/queue for every reply. This behavior was actually introduced in grizzly, but is off by default. If you set amqp_rpc_ single_ reply_queue= True in your config file, you should not be seeing this problem.