That is, imagine that after a version of the data is released it increases 10X, but the schema doesn't change. That might still be worth making a release for, right?
Typically what I see in the wild are data dumps on some sort of semi-regular schedule such as weekly, monthly, or annually, "versioned" with the timestamp or some derivative thereof (eg. "August 2016").
This is much better than nothing, but just as a source dump for each release of an application or other software in a tarball is not as useful as being able to check out a tagged release from a repo, so too versioned data releases that can be checked out with a tag from a repo would be more helpful that a data tarball.