Could the "vagary" be that the only instrument approaches to the airport are routed through there.
Bad word choice, I should have said "arcane factors".
In this particular example, there are approaches from three directions at ADM - runway 13 (the extension to the NW), a VOR-A from the southwest that is aligned with the (now closed) runway (4?), and one to runway 31 from the southeast (where there is no extension).
Notice that the whole area is within the shaded magenta indicating Class E at 700 AGL. Also, in airspace design, a 300 ft buffer is applied to this, so effectively an aircraft at 1000 AGL is considered to be exiting this Class E.
The FAF for the RNAV (GPS) RWY 13 (6.2nm final) is far enough outside the Class D that a descending aircraft could descend below the 700' Class E prior to entering the Class D - but not by much, so it's a short extension.
The FAF for the VOR-A is well outside the Class D (it's an 8.8nm final), so the extension has to be longer to contain the aircraft within Class E.
The approaches for runway 31 (GPS and ILS) have a FAF that is 5.7 and 5.4 nm long, respectively, so the aircraft doesn't descend below 700' AGL until within the Class D, so never breaks out the bottom of Class E.
Airspace boundary calculation is something that AeroNav Products does with every instrument approach, then the information is turned over to the airspace people. It relies on steeper than normal descent gradients to account for varying pilot technique.
This is covered in more detail than you'd ever want to know in FAAO 8260.19F, chapter 5, including how it's calculated for stuff like procedure turns that don't have a FAF.