The short answer is a matter of the size of our testing instances, they are pretty small, 5 to 10 GB. This is fine for modest projects and wonderful for our preferred cloud-centric, CI workflow.
Until a project database and accompanying blobs start to get bigger than 10 GB. Then the 'representative subset' approach starts to become very, very, very appealing. In fact, even if we were to test on the full database at critical points, there are many scenarios where the 'representative subset' is not only adequate for testing and development but far more efficient.
Even if you disregard our cloud based 5GB testing instances, think of the inconvenience of having to pull down 800GB to a local development machine.
(Maybe these are small project, modest budget, problems ).