It could not be more obvious that a single test would be scientifically and statistically invalid.
A large number of repetitions would significantly reduce random differences affecting plant performance and characteristics, but they will not be eliminated. By increasing soil uniformity, mapping soil conditions (and taking them into account), and repeating the tests multiple times over many years, the accuracy of the tests can be significantly increased-- but there will always be randomness and minute variations that cast a shadow of doubt over the results. Over time analysis of an increasingly large data set should be able to generate a "preponderance of evidence" for various conclusions...