So the issue is either that the .0004 second improvement in the operation represents more than a 25% optimization (since an actual 25% optimization should result in a linear speed gain), or that the optimization of a single item somehow resulted in a non-linear performance improvement.

That, I can't help you with; perhaps dws is onto something. I couldn't run the benchmark and haven't the time to tweak it and break out "get" versus "set", at the moment. (Nor have I the time to speculate on why it might not be a linear improvement.)

Apparently, I've got just enough time to be no help at all. At least my keen grasp of the obvious is intact.