It is rarely a good idea to bring legal analogies, but I consider this to be one of those rare cases. Notice, I am in no way implying any particular legal connection of falsifying data to particular crimes.
Let's consider a typical US criminal law, and take Connecticut as an example. Connecticut's laws define various crimes which can be of different classes (A–D, unclassified). Class A crimes constitute the more serious types of crimes and carry longer sentences and large fines, while Class D crimes are the least serious kind. For example, Class A felonies can lead to a life sentence, while Class D felonies can lead to a small fine. However, no matter what class does the crime belongs, it is still a crime.
Similarly, data should never be falsified. However, the seriousness of the offense and the consequences (in case, it is discovered) would differ. The impact and, maybe even more importantly, mens rea (guilty mind) of the falsification from a first-year B.Sc. student is different from an established researcher.
Moreover, one could argue that more serious crimes usually originate from less serious ones (discovered or undiscovered); therefore, it is extremely important to have students educated about the Code of Honor and the importance of research\educational integrity as early as possible.
To summarize, falsification of data is always really-really bad. The severity of the consequences might vary. But, as it is with any criminal record, it makes it very hard to overcome this event in the future, potentially closing some roads forever. And even if not caught, this might lead a student onto a dangerous road, and I would strongly advise against it, no matter how much immediate benefits such data falsification might offer in a moment.