I expected much better from the Register. The salient points from the article that got cribbed from were completely missed.
1: Most of the meters tested WEREN'T smart meters (but the ones that were, were just as inaccurate
2: the 2 under-reporting meters were the only ones using Hall Effect sensors.
3: All the rest overreported
4: The grossly overreporting ones used Rogowski coils
5: ALL the meters were accurate with purely resistive or reactive loads (capacitance or inductance)
6: Complex loads are the problem:
The problem arises when switchmode PSUs comprise most of the load - their spiky current draw (the input side is rectified to DC and fed through into a storage capacitor - current is only drawn when the rectified DC voltage exceeds the capacitor's voltage, leading to current usually only being drawn through the top 10-15% of the voltage curve.
The meters interpret this as a _much_ bigger equivalent resistive load
These kinds of current draws are a bigger and bigger problem on grids. They strain distribution transformers (current capacity (ie VA) needs to be larger for any given wattage rating) and frequently cause high currents to flow in the neutral line in 3-phase systems (ie, your street main) and in some cases have caused DC offsets to earth, which _really_ messes up transformers.
Whilst there have been rules about eliminating this kind of current draw on PCs and other such PSUs for years (mainly because of the problems caused in office buildings by lots of the things - they can trip out ELCBs as one example of nuisance value), smaller domestic kit hasn't been required to have correction circuitry built in - as the majority of domestic load these days is lighting, the tests were done using CFL and LED lamps as the loads to see how accurate the meters were under such conditions. Switchmode PSUs are in everything these days, so it's quite possible the errors are significant in £££ terms (having it indicate you drew 5kWh in lighting over a month isn't a big error on 1kWh given the cost per kWh)
On the other hand (and not mentioned in the original article) the EU is harmonising rules on all switchmode PSUs requiring their current draw is more-or-less in phase and non-complex, so this problem might go away soon anyway.