I think if model 3 will be 20% smaller than model S... the weight will be at least 20% less (at least, but maybe even less due to smaller battery pack)

I'm not so sure about that. Remember, they've talked about moving to more steel and less aluminum.

and the drag will be 20% less too (same Cd with lower front surface).

If they were to just talk the Model S and scale it down to .80 scale (which they've said they won't), each dimension would only be scaled down to the cube root of .80, and the front surface area would be scaled down to the square of the cube root of .80, which is ~.86, so the drag would be ~14% lower. On the other hand, it's likely that the height won't be scaled down quite so much in order to leave acceptable headroom, and the length would be shortened more to make up for that. In the that case, the frontal surface area might be .88 or even .90 that of the Model S. That being said, there's also talk that the Cd will go down as far as .20; I doubt that, I think .22 is a little more likely.

This means than model 3 will need at least 20% less energy for the same range than model S.

According to my calculations, it'd be more like 15% less energy.

I bet lower cost Model 3 (35000$) will have around 50Kwh for 200 miles EPA

Elon has said over and over again that 200 miles real world range is the minimum acceptable level for an electric car, and that in order to do that you need significantly more EPA range, something like 220-250.

(the model S60 gets 208 from the 60KWh pack)

208 miles at 20% less energy and 5/6th battery size would be 216.7 miles, not 200. Under my estimates, a model 3 50 would get 204. Again, I don't think Tesla will release a base 3 without 200 miles real-world range, which means a bare minimum of 220 miles EPA. In my mind, that means a 60 kWh base.

I also bet for higher range model 3 with a 70Kwh battery pack and a similar range to a 85/90KWh model S (20% of 85 is 68).

Others improvements are also possible so the range could be even better with these battery packs.

I definitely agree that there will be a 70 or 75, and perhaps also an 85 or 90. After all, JB Straubel did say that the third generation (3/Y) would have about 40% better volumetric efficiency than the second generation (S/X).