It is very unlikely the 10 (or 20 or 30) second sub-exposures are exactly that, continuous exposures of that length. If the telescope does no tracking during the exposure, the image can move up to 15 arcseconds/second, depending on declination. Since a pixel is 2.44 arcseconds, that means up to 6 pixels/second, or 60 pixels for a 10 second exposure.
Star images on the S50 are very sharp, fainter stars can be only 4-5 pixels across and are perfectly round, showing no movement or trails. Since tracking during exposure with this kind of precision is completely out of question, I think this means the exposures are much shorter, probably under one second and they are stacked together electronically to create a "10-second sub-exposure", especially since with the mount not moving you can predict the image shift for each sub-sub-exposure and compensate for it by shifting the image. This is further confirmed by inspecting the 10-second consecutive subs, they are constantly shifting with respect to each other, and every few subs or so there is a bigger jump back to compensate for the accumulated shifts and roughly track the target. The mount moves probably once every minute or so and never during exposures and the final image is a stack of 10-second sub-exposures while each sub itself is a stack of a few much shorter exposures, definitely sub-second ones.
Stars of magnitude 12 are clearly visible in "single" 10-second exposures and they are perfectly round and just a few pixels across. In a few minutes long "stacked" final images I was able to identify stars down to magnitude 15, from a Bortle 8 zone!
So in conclusion there should be no major difference between a 30-second sub and three 10-second ones stacked together, the real individual exposures must be much shorter anyway. The only advantage I see for 20 and 30-second long subs is that you save on disk space for the same total exposure time. So I do not think there is a real benefit of 20 and 30-second subs, on the contrary, the longer ones increase the risk of accidental movement and losing the entire exposure. Which by the way, happens a lot, I see a large percentage of dropped 10-second subs, between 25% and 50%. I did exposures as long as one hour and they took more like two hours to complete. This is something ZWO needs to fix as it is a terrible waste of time. Not to mention that there is no automatic way to end an acquisition and save the result, you have to be there and press the stop button in person, otherwise you get not image unless you also save the subs.
There is no fundamental reason why you could not say for example "take a 5-minute exposure and save the final result even if I am not there to press the stop button". Or even better, save a "final image" every minute or some other programmable interval, a FITS image is only 4MB and the JPGs are even smaller.
The Seestar app is incredibly limited in functionality and is not at the level of the S50 hardware - it looks very slick but lacks major features which would be trivial to implement. For example the annotation overlay, which is very useful but you might not want to have it in the final JPG image, why not save it separately in the same acquisition folder and let the user decide later whether they need it or not in their final image. More star labels are also required in that overlay too.
And my most important complaint of all, how long do we have to wait for a panoramic mode, which would stitch together multiple frames into a single much larger image. This feature, especially with some simple automation like "take as many 5-minute unattended exposures to cover a 4 degree by 3 degree area centered on M31" would be awesome and make the S50 even more impressive and useful than it already is. If you tried to take a full image of M31 or M42 or whatever, things that do not fit in a single 0.72x1.28 degree frame, which you cannot even rotate, you know what I mean. Plan a 4 or 5 series of 10-minute frame captures, all while the sky moves and rotates and the S50 drops 50% of the subs and it can quickly turn into a really frustrating exercise and a huge waste of time. What is even more annoying is that the hardware is there and could do all this and much more, but the software is lacking.