Why use a (EIP) Normalizer instead of keeping a separate queue for each data format?

Within the context of enterprise integration patterns (EIP) there is the concept of a Normalizer -- which is composed of a queue, a content based message router, and a message translator to translate the different data formats to a uniform one.

I have always kept one queue for each kind of data. So when is this pattern necessary? It seems better to have a separate queue for each data format and route them directly to the appropriate translator -- and not have to rely upon (probably brittle) message identification.

Am I thinking about this wrong?