Then one makes the judgement call:

    cost of testing w/ production-like data > cost of potential production issues

and decide whether it's prudent to omit or scale down the tests.