QA was probably done on linear loads which is where the meters would be accurate. On noisy switch mode loads (VSDs, computer power supplies, etc) there was probably minimal QA. Quite likely because regulation has not caught up with the prevalence of these loads. It is a case of meter manufacturers not keeping up with the time. And yes, I have seen this on a large scale; I ran into one power meter with a 200kW difference from the 1MW VSD it was down-stream of. And that was with some harmonic filters installed on the drive, yet the meter still read less than the drive. I'm inclined to suspect the meter here.