Is there a standard in implementing MultiPV in regular engines?

[Laskos]

You mean all the other moves have to be searched only once, rather than N times? But it is not the same search. Normally they would only have to be proven worse than the PV move, but in multi-PV they would have to be proven worse than the Nth-best move. Which has a lower score, so that this requires a better refutation, which would take more time to find.

You almost convinced me that my stupid intuition was stupid indeed, but now, correcting for time control issues (playing at fixed time per move), it seems that most top engines (3 out of 4) have a MultiPV which saves a factor of about 2x compared to 4x time in MultiPV=4 to the similar quality best move. So, after all, SF and other do follow my naive intuition that 4x MultiPV=4 ("well" implemented) should be better than 1x MultiPV=1 as best move choice goes. And only Komodo implementation of MultiPV follows weirder rules (even more than 4x time is needed in MultiPV=4).