I’ve observed that with very heavy JSON or csv files (thousands of lines) importing and adding to a comp can take a very long time. (For example a multiyear multiregion COVID dataset, or huge GPS tracking data)
In those cases parsing the data myself (either naively using .split() methods or with an external library like papaparse converted to work in expressions) was much faster.
I doubt that there’s a lot of impact on smaller datasets with just a few colors and texts etc.
Adding a data object to a composition actually just prefills everything with expressions that fill in the right coordinates in the dataValue() method, but it is a convenient way to inspect the structure of the data, get the number of rows, and to get the column names etc.