Reusability of data is an important part of research. It helps collaboration between researchers, and enables secondary research to take place. Using the same data is important for reproducibility in many cases, because the research isn't about creating the dataset, it's about doing analysis on the data. A lot of original research relies on existing datasets.
Having "good" data is obviously crucial, but it's a separate matter.
It’s not a question of “good” data. Slice and dice perfectly random data and sometimes you get spurious correlations. The only way to separate them from real results is to have completely new data.
It’s not even a question of p hacking or bad design. Preform enough experiments and you always get false positives.
Suppose two people conduct the same experiment on the same medical data using the same code. If the sample was biased then what’s the point?