Here is the statistical table for a 2 tailed paired forced choice preference test, which is what you are planning if you give 2 different cigars to people and ask which they prefer (and don't take no preference as an answer).
As you can see, it is not vaid to run the experiment with 6 of fewer observations and that you need more than 8 observations so you don't require 100% agreement. We never run these tests with less than 72 observations if the data are important. But for the sake of argument, if you have 25 observations (i.e., 25 different people smoked the cigars and gave you their preference), 18 of them would have to prefer one of the 2 smokes for the difference to be significant at the 5% level.
And you would need to run the experiment using a number of different comparisons.
I did something similar when I ran an accelerated aging experiment several years ago. I asked people to smoke the cigars side by side, and compare them as the burned down. I did learn that you need to find people who are motivated and who have palates that you can trust.