>The requirements were quite specific so I'll have to verify to know why they want it the other way. Basically, if no error will be generated on the client side when sending such an XML string, that means we can basically sent whatever we want here, as described in the requirements, correct?

Yes.

>So, this approach to send errors is basically to send something that will be easy to interpret from the client side as oppose to use COMRETURNERROR(), correct?

Yes, that's the purpose of it. COMRETURNERROR(), in this context, has the same effect as a severe communication failure.