Cheating is unfair regardless of how you handle grading, but how unfair it is depends on the grading scheme. Let's assume that the cheater's efforts earn him among the highest raw scores in the class.
No matter how you adjust the grades (unless you flip high and low grades!) the cheater will have a higher grade that they deserve, and anything which is grade dependent and competitive (e.g. getting a job, getting into grad school, etc.) will be unfairly easier for the cheater and unfairly harder for everyone else. So there's a baseline unfairness to cheating.
If, additionally, you keep the percentage of A's fixed (grading on a curve), the cheater will bump someone out of an A, which will likely impact them negatively (career, grad school, etc.). Someone else may be bumped out of a B, etc, down to where the cheater would have been if they hadn't cheated. So there is direct and noticeable harm to the cheater's fellow classmates. It's a pretty rotten thing to do. (This is also related to why students fight so ardently for every grade or grade step.)
If you keep the requirements fixed, then everyone will get the same grades they would have, save the cheater, but the average fraction of A's will go up. The net impact is the same--a 3.2 average now looks worse than it did before because the averages have gone up--but the negative effects are not felt by individual classmates but rather averaged over everyone from the university with cheaters. By spreading the harm among more people, in some ways the effects are less acute; aside from the cheater, everyone else is still going to have the same relative standing.
So, especially if cheating is rare, it's even more unfair if you're grading on a curve* since those people who have the bad luck to be stuck with a cheater will be unfairly punished in addition to the cheater being unfairly rewarded.
* There is an argument for grading on a curve, however: there is also unfairness due to differences in grading schemes between universities. Grading on a curve helps to combat this, as instead of measuring various different expectations of mastery (e.g. if Harvard expects you to do more to get an A than does Yale, then your GPA will be lower if you go to Harvard than Yale), it just measures students against each other, and there are enough of those for statistical sampling help even out the other differences (e.g. if the incoming classes to Harvard and Yale are of similar quality, and both grade on similar curves, regardless of which one you'll know who was in the top 10%, and those top 10% will be similar regardless of where they went).
