CouchDB is more C++ than Erlang these days, bad example. It hasn't been mostly Erlang for a really long time. Erlang basically only does the clustering behavior, which is something Zookeeper et al. do already.
Riak is the only solid example out of those you listed, and its known for having poor raw performance, excellent availability, easy clustering, and awful usability.
Interesting, where can I learn more about this claim? What are the incompatible facets of a generic queue and Erlang? What is an example of these generic queues (or if it's only an ideal, what are its characteristics)?
If the queue gets badly backed up, you're probably fucked.
Larger, longer-lived work queues are really more of a Hadoop thing, not an in-memory queue thing.
Where the first answer states that: "RabbitMQ presumes that consumers are mostly online, and any messages "in wait" (persistent or not) are held opaquely (i.e. no cursor). RabbitMQ pre-2.0 (2010) would fall over if your consumers were too slow, but now it's robust for online and batch consumers - but clearly large amounts of persistent messages sitting in the broker was not the main design case for AMQP in general."
(It's contrasted with Kafka, which is "designed for holding and distributing large volumes of messages")
Still, I see no reason why you say that it's Erlang that causes this?
You can implement a queue that works better for batch work in Erlang, it's just that it'd be closer to a data store.
Isn't that true of any queue ?
 Or at least Joe Armstrong says so in his book, "Programming Erlang".
Can you point to any references?