When looking at terciles, quartiles, quintiles, sextiles, etc., it is important to know the usage range associated with each group. If the amount used/consumed/performed by neighbouring groups does not differ significantly, the division serves no useful purpose, except possibly to mislead. The N of the study needs to be high, and there needs to be a significant difference in usage/consumption/performance levels for the division to be meaningful.
The technique is simple, you rank your N participants in decreasing order of carb consumption, start at the bottom and take N/5 participants, then take the next N/5 participants, and so forth. If there are a lot of people at the same consumption level, you may have to assign them at random into one quintile or the other. You can then characterise the groups by their mean, mode, and median consumption and sometimes get some interesting results. But if, say, out of a study of 100 people, the bottom 20 have 15 people who ate 50 g of carbs and the next 20 have 15 people who ate 55 g of carbs, the inferences you can draw are not very significant. So I would treat this study’s conclusions with great caution, unless they give all the details involved.
In the College Board and SAT exams, for example, the population taking the test is numerous enough, and the variation in scores is wide enough, to make it worthwhile dividing the test-takers into percentiles.