Given full stoichiometric burn, you'll want droplets small enough to burn fast at ignition time, distributed evenly across the chamber. No droplets at all makes for a gas mixture that is harder to ignite and keep burning in the initial stage after spark. Remember, we want burn, not detonation. If only very few fuel atoms are burning, the flame might extinguish if our spark wasn't long en powerful enough. Even if it doesn't go out, it will take time before the chain reaction of burning gets momentum. Once you have a proper flame front going, you may want other conditions to apply than at ignition time.
If you have minute droplets, each burning droplet will have it's own fuel supply and there will be plenty of chances for it to set fire to it's neighbors. If you have too large droplets, they will take a lot of time burning, possibly even run out of oxygen and not burn completely.
Before you ignite, it doesn't really matter how big the droplets are, as long as they are evenly distributed and in suspension. You might benefit from the cooling effect of the vaporizing of the fuel on the air you ingest.
What is the exact "best" droplet size for that time, engine and application changes every millionth of a second. It's a very dynamic process in the combustion chamber and even the most advanced computer simulations (think formula 1 multi-million dollar setups with software that costs just as much) can only model it up to a certain extent. Once you set fire to the fuel, pressure, temperature, shock waves, reflections, resonance, chemical composition all influence the burn process. Even if you divide the chamber into small 3D cells, you can only calculate a very small time frame of what is happening per cell. All these cells influence each other. All cells are only valid for that time frame, since the piston moves. The next time frame, you have to re-define your cell layout, put in all data you got from the previous time frame (try extrapolating values for a few million cells) and start all your calculations again.
The dynamics and complexity of the model is easily larger than that of the Met offices computers. Those use a 3D cell model as well and use external input variables like pressure, solar power, cloud cover, resulting wind speeds, ground temperature and such. In a way, they can be compared to a model of a combustion cycle. Try calculating the weather for over a week, in 15 minute interval time slots and see how accurate you are after a week. I've worked for the Dutch Met office and believe me, it takes hours to do a weeks run for just the country and you only get to maybe 50 percent accuracy for the 7th day. That's 700 time slots for one week. If you try to do an entire combustion event in just 700 time slots, you'll never get the accuracy you need to calculate things like ideal droplet size. Even if you'd do 10 times as much, you'll only get droplet size for one very specific load point at one RPM number, engine temperature, barometric pressure, air temperature and so on. Coming up with a full range data calculation for just one design of engine is not practically possible, let alone for all possible variations on that engine. Currently, F1 is picking specific situations for partial calculations using models like this. Also they are doing a different modeling called Fluid Dynamics, where they treat gasses like a liquid and compute the flow of the "liquid".
It's a long post, but I hope it's clear now that "you want the fuel to be atomized completely" does not hold up. Gasoline is a complex mixture of hundreds of different substances that all have their role in the burning process. You don't want a violent explosion, where you'll convert fuel and air into a hot gas at a rate of hundreds of meters per second. If you do that, you'll shatter your piston and be left with nothing but heat and no energy to move your crank around. Getting the correct pressure build up requires careful burning of the mixture and thus a careful preparation of that mixture.