Yes, that see-saw effect is one reason I remain unconvinced by metrics like “long term agreement with revenue meter to within 1%” especially for per-breaker monitors. That all the individual channels add up to within 1% of the revenue meter over a month doesn’t really put any upper limit on the errors in individual instantaneous power readings from each channel, and surely the whole reason for per-breaker monitoring is so you can determine how much energy circuit x uses.
And in three phase installations, per circuit monitoring with “derived voltage references” (i.e. only one VT) is particularly vulnerable. In that situation I think the “long term agreement with revenue meter to within 1%”, while no doubt true, is close to meaningless for anything but the grand totals. But if that’s the only measurement you can rely on, you could have saved yourself a lot of CTs and just measured the main feed.