Caching could be more efficient, easier to automate than scripting, if users could create a feature class that could control both extent and LOD for a single cache job.
For example, a feature class could contain several features of various extents, with certain attribution that server could use to control which feature is used for specific LOD's. Users could then initiate a single cache job, with a single feature class. This would eliminate the human from the loop, minimize server idle time, and keep users from having to initiate sequential cache jobs.
The feature class could look something like:
OID MinLOD MaxLOD
1 L00 L07
2 L07 L10
3 L11 L16