Is it possible to somehow isolate the growing units in a way that's both economical, compatible with a degree of automation and also limits cascading failures like disease?
It might be. But that's the whole point of the objection - we suspect that naive scaling is unlikely to scale in the expected manner. There might be ways to mitigate that scaling losses, but it's disappointing that there's no investigation (or mention) of scaling losses in the paper.
Light, nutrient and water distribution can be replicated, but making sure things scale properly and you don't get cascading failures is key.