Since Windows Sockets applications can't possibly be aware of what type of remote computer system that they will be dealing with a priori, it is necessary to define a common data representation model for vital information. Windows Sockets chose the big-endian model for the "on-the-wire" data representation, known as network byte order.
The Windows Sockets interface offers APIs to application programmers to do the necessary conversion between the local system representation (or host byte order) and network byte order. There is no harm in using these routines on systems which store data in big-endian natively; in fact, it's encouraged. By religiously using the byte-ordering APIs, your code can be used on systems with different internal representations without inheriting byte-ordering problems, thereby making your code more portable.