Based on my experience teaching I don't think individual variability is that large. In a semester-long course with 3 or 4 exams the top students and bottom students are mostly consistent from exam to exam.
That said, of course individual variability does exist. In one graduate-level math class I misremembered one theorem which caused me to immediately lose 40/100 points--there were only 5 questions on the exam, and that theorem was central to two of them. Had those two questions been different I probably would have scored much better. I consider that a poorly-written exam, though, since so much of the score was dependent on recalling one theorem correctly. I somehow still ended up with an A in that course, I can only assume through either creative accounting or a generous grading curve.
In any case, I don't know any other method of evaluation which is more fair than exams. Every method of evaluation is subject to similar sources of individual variability and some have other issues in addition. "Fair" in such cases is a mirage. It's not like there's some other objective evaluation that we can use instead of exams, just ones with different biases.
That said, of course individual variability does exist. In one graduate-level math class I misremembered one theorem which caused me to immediately lose 40/100 points--there were only 5 questions on the exam, and that theorem was central to two of them. Had those two questions been different I probably would have scored much better. I consider that a poorly-written exam, though, since so much of the score was dependent on recalling one theorem correctly. I somehow still ended up with an A in that course, I can only assume through either creative accounting or a generous grading curve.
In any case, I don't know any other method of evaluation which is more fair than exams. Every method of evaluation is subject to similar sources of individual variability and some have other issues in addition. "Fair" in such cases is a mirage. It's not like there's some other objective evaluation that we can use instead of exams, just ones with different biases.