Empty or full
Well, actually it is more likely to achieve the claimed PUE when it is full as data centre PUE tends to get better with load, meaning a data centre that is only part full is likely to have a much worse PUE than when it is at full load.
The adiabatially assisted economised cooling is certainly not very common but CG are far from being first to use this method to cool buildings in a Swindon like climate, as for complex controls, those make me nervous, give me simple and reliable controls any day over complex and unpredictable failure modes.
What will be interesting is to see what PUE it achieves later on when some customers have demanded 2N UPS and various other efficiency impacting nasties. It is quite easy to get down to 1.1 or so with this type of cooling so long as you don't have much resilience and rely on the cloudyness of the IT platform to provide your failover.
Of course, this assumes they are including all the shared infrastructure in the PUE calculation and not telling porky pies like certain vendors of container data centres who leave out all the important bits like the transformers and UPS when claiming "can achieve a PUE of..."