I tried it last night. The difference was frankly pretty astonishing. It was too windy for any meaningful test with imaging, but I did some somewhat qualitative testing between gusts since multi-star guiding has no particular advantage when dealing with wind.
The idea behind multi-star guiding is that the majority of seeing effects are localized to areas on the order of arcseconds in apparent angular diameter. This means that each star that is any meaningful distance away from another star in the guide field will experience completely different seeing. Since seeing results in random displacement of star images, averaging the displacement of multiple stars will cause the displacement to add in quadrature just like noise does. PHD2 uses (up to) 9 stars for its multi-star algorithm so we can expect to see "seeing" effects reduces by a factor of the square root of 9, or 3.
My test was to run PHD2 unguided and observe the Declination RMS deviation for both the single star and multi-star algorithms. I selected Dec to remove tracking error as a variable. If theory holds, the RMS deviation in Dec should be about 3 times higher for single star guiding compared to multi-star guiding. I found that using single star guiding, my Dec RMS was about 0.45". Using multi-star, it was about 0.21" for a difference of a factor of 2.1. Not quite the 3 that theory predicts, but probably within the error bars of my subjective test. I say "subjective" because I had to reject data that to my eye appeared to be the result of wind gusts. I look forward to trying this on a calmer night.
I will say that the effect was pretty remarkable. It was literally like flipping a switch as I toggled multi-star on and off. With it on, the graph would turn to a beautiful near-flat trace, and with it off, the trace would jump around quite a bit more. As for actual guiding, I only caught glimpses between puffs of wind, but during particularly gentle times, I saw patches with each axis less than 0.3" RMS and Dec often dropping below 0.2". My typical guiding performance is around 0.5" per axis, so that is a dramatic improvement.
One thing I did notice is that the multi-star algorithm reduced seeing effects enough to reveal the roughness in my mount's RA tracking. There would be times for minutes on end between gusts where Dec would be below 0.2" RMS, and even periods when it dropped below 0.1" for meaningful periods of time. But my RA seldom dropped below 0.25". So it appears that my mount is just not capable of tracking in RA below 0.25" RMS. That's not a complaint, just an observation. Tracking at 0.25" RMS is actually pretty remarkable.
Anyway, my preliminary thoughts are that multi-star guiding is a huge step forward for guiding.
If the test was run with PHD not guiding, doesn’t that just show that with no actual improvement in physical guiding, the multi star just calculates a lower RMS than single star. You could therefore reasonably expect that when guiding is enabled you would still see an improvement in calculated RMS but the physical guiding may actually be no better.