Is it worth designing a system to expect test accounts and products to be present and active in production, or should there be no contamination of production databases with test entities, even if your shipping crew knows not to ship any box addressed to "Test Customer"?
I've implemented messaging protocols that have a test="True" attribute in the spec, and wondered if a modern schema should include metadata for tagging orders, accounts, transactions, etc. as test entities that get processed just like any other entity--but just short of the point where money gets spent. Ie: it fakes charging an imaginary credit card and fakes the shipment of a package.
This isn't expected to be a substitute for a fully separated testing, development, and QA database, but even with those, we've always had the well-known Test SKU and Test Customer in the production system. Harmless?
Having testing accounts in production is something I usually frown upon because it opens up a potential security hole. One should strive to duplicate as much of the production environment in testing as possible but there are obviously cases where that isn't possible. Expensive production only hardware is a prime example. I would say as a general practice it should be discouraged but as with all things if you can provide a reason which makes sense to you then you might overlook a hard and fast rule.