The idea behind "multi-star guiding" is to use the centroids of multiple stars instead of the centroid of just one star as a reference signal to tell how much the mount is deviating from perfectly tracking the sky.
When atmospheric turbulence is bad ("seeing" is poor), the stars themselves would appear to be moving noisily. To avoid mistaking the position of the guide star that is caused by "seeing" as an actual error of the mount, one way is to average out the stars for a longer exposure time.
If the "seeing" error is random, then each time you double the exposure time, the variance of the "seeing" error drops by a factor of two (i.e., the standard deviation, or RMS error, drops by the square root of two).
There are multiple reasons why using a longer exposure time is bad. The long exposure will bloat the stars, for example, and that alone will make centroid finding more error prone. Another has to do with the stability of a discrete feedback loop when there is significant latency. You have probably noticed in the past that longer exposures can tolerate less guiding "aggressiveness."
Fortunately, probability theory tells us that you can also reduce the variance of the error by using multiple samples, as long as they are independent. (This is known as stationary ergodic processes.)
If you were to double the number of samples and they are independent, again, the variance of the error also drops by a factor of two.
This make it very nice for autoguiding, since every star is affected by a different pocket of the atmosphere. In a dark night, you will see that the stars twinkle independently, instead of in unison.
So, using the centroid of two stars that are far enough away from one another, a one second exposure is the equivalent of using the centroid of one star in a two second exposure. Using 1 second exposure of 4 stars would reduce the RMS error of the measurement by 2, and thus the equivalent of exposing for 4 seconds. If you guide with 12 stars, all of which have the same signal to noise ratio, a one second multiple star guiding is equivalent to using a whopping 12 second exposure.
The centroid errors are usually weighted, and in the case of ASIAIR, by the signal to noise ratio (SNR) of each star. Because of that, in general, 8 stars will not give you the full force of 2.8 times lower measurement RMS. Unlike in the single star case, where an 8 second exposure will give you a 2.8 times lower measurement RMS error, since the single star's SNR is not changing by much over the 8 seconds.
There are other problems too, such as the need to take into account field rotation over the duration of multiple subframes, etc, and if clouds have obstructed one or more stars.
ZWO decided to use a star with the best SNR as a primary star (this is the one that is currently inscribed in a square) under the green cross hair. It is probably to make their computation easier, even though there are disadvantages to having a primary star. When that star is lost due to a passing cloud (or even an owl), guiding is also lost, for example, until that single star is back again.
All the other stars will also drop out and come back. However, each time you lose a non-primary star, the measurement error only goes up slightly (because of fewer centroids), and is usually is a non problem.
Because of all of this, it is crucial that even dimmer stars of the 12 possible ones all have good SNR (or you will just be weighed predominately with just the centroid of one star). This means that your guide scope focus must be tack sharp. Things like pixel scale becomes just as important as the choice of the pixel scale of your main OTA.
You will need to also polar align a bit better unless ZWO has subsequently solved that problem. With single star guiding, the guide star will stay put while other stars revolve around it, so field rotation is not as critical to autoguiding. With multi-star guiding, the pattern of all 12 stars need to be fixed relative to one another.
As time goes by, and if ZWO adds fixes to the potential secondary problems, multi-star guiding should keep improving.
There is no magic with multi-star guiding, it is just the use of probability theory to mitigate the "seeing" problem without having to increase guide exposure times.
Chen
