This could be different from what is under discussion here, but I've
seen "false levels" at multiple institutions while cleaning up batches
of EADs. As far as I can tell, the main reason component padding has
been done historically is for display purposes. Some processors in the
past may not have understoond CSS or didn't have access to the finding
aid CSS files.
For us, it was not a matter of CSS padding, it was problems that occurred during XSL transformation. The XSLT we were using was a tweaked EAD cookbook stylesheet that couldn't handle a collection that had series with sub-series and series without sub-series. For example, we were working with many collections intellectually structured like:
c01 Series: Vessel Papers
c02 Sub-Series: M/V "O.K. Service X"
c03 File: Manifests
c03 File: Captain's Logs
c01 Series: Financial Records
c02 File: General Ledger
c02 File: Annual Reports
As I understood the problem, the XSLT we were using was not testing individual components to see if the attribute was series, sub-series, file, etc. It thought all c02s were sub-series and would not correctly transform c02s that were files. That might be a simple fix, but it was beyond our XSLT knowledge. In the absence of a dedicated XSLT person to write the XSL test for us, we added "false" c02 layers to our series that did not contain sub-series so all files would be encoded as c03s. We then uploaded static HTML files into the university's content management system, which controlled all the CSS.
It was less than ideal, but it got the job done. I am especially interested in Ethan's comment about just using <c> and @level. Is anyone else doing this?