The LEGO block analogy for SOA is wellknown and the message is that a service should be like a LEGO block: well-defined interfaces, reusability, easy to assemble into new compositions, orchestrations and mashups; but also a governance problem with all the different services floating around; and add that even if it is easy to compose new structures from Lego blocks, it might not be that simple to change Lego systems.
Another analogy that is useful when discussing service-orientation is Darwinism (Charles Darwin) - more specifically that specialization of a species to a specific habitat makes the species less adaptable and more vulnerable to changes in their environment. If you think of a service as a 'species' and of the 'habitat' as the service context, you'll see that a service that has a very high coupling to its context is not very well suited to being reused in another context. The service is just not agile. The service is just not ready for SOA bliss.
Just as if you relocated e.g. a moose to Africa, it would die as it is specialized to the climate and the type of food (birch/pine) of its habitat; a service that depends on e.g. the employee directory cannot easily be reused on your company's public web-site.
I think that the agility of a service is a good indication of whether your services are truly SOA rather than plain JBOWS. Providing good SOA services is more difficult than you think, making just a web-service is far to easy. This is the hardest message to get across to JBOWS developers claiming to provide SOA services. Afterall, a web-service that is specialized to fit perfectly to e.g. the current intranet application will become extinct when the business environment changes. And nothing changes faster than business. A moose in Africa stands a better chance.
Evolve, as natural selection will see to that repeating the failure of DCOM/CORBA for distributed systems using web-services has no future in SOA.
Btw, Dion Hinchcliffe has a good article about SOA, WOA and SaaS in 2007, referring to the web as a Darwinistic 'petri dish' software experiment that will show us what works.