No you don't. You do need a properly randomised sample though. Asian people are generally shorter than Caucasian, and people from North Korea even more so, so a sample which was not properly random (say taking 30 men from North Korea and 30 women from rural Germany) would be misleading. But a sample of 30 random people across the entire planet would not, even if we didn't control for height. Weighting for regional height differences would give us a better understanding, and increase our statistical power, but you do not need to control for all variables, and in fact in most research controlling for all variables is impossible. Even in something like basic physics what you are suggesting is impossible. All g-2 measurements for the electron for instance have been conducted on Earth. We have no way of knowing if that works outside of a stars gravity field or in a black hole. It is still reasonable to assume it is.
The sample in climate research is adequately controlled, sufficiently random and the noise is well enough understood that it is likely most of the conclusions of the research are useful and predictive.