The only thing i dont like about this is having the xpath expression hard coded in the orchestration. Ive seen this a number of times where the schema then changes and you end up with a strange defect somewhere later in the process and you end up tracing it back to the xpath expression being wrong. I tend to prefer the xpath expression being defined in a constant in a C# library which is then tested against a message instance as part of a unit test. The benefit of this is if the expression becomes invalid you would know much quicker.

In what we have done this benefit outweighed the additional complexity to the deployment/solution