best statistical method to calculate statistical significance between columns

This is a tough question. In part I think this is because none of us here seems to be a good enough statistician to really be able to advise you on your choice of statistical test. A couple of question to help us understand what you are doing:

1) How do you expect the attempted t-test to account for total orders? Other than being used as the divisor in calculating the percentages, a t-test comparing the difference between these two averages is not going to otherwise account for differences in the total order. Perhaps you need to think more carefully through how the percentage is calculated to see if there is a different way to do it to better account for the differences in total orders. Calculating percentages is a common way of trying to account for differences in totals, by "normalizing" both scenarios to the same scale, but maybe that isn't what you need to do here. Or maybe you need to think through more carefully what these percentages really mean to you.

2) What does average percentage mean in these cases? I know how to arithmetically calculate the average percentage for these four fruits, but I don't quite see what that average would tell us. In addition, I'm not sure I understand what it means when we compare one average that is heavily skewed towards oranges with another average that is somewhat skewed towards grapefruit. It just isn't clear to me what difference(s) we are looking for in the two data sets.