As reported by forum user Bomoo and confirmed by my own testing:

The decoupler in the Avionics Ring appears to separate from the stock Mk1-2 Command Pod with greater force than the Service Module, even though both have ejectionForce = 200.

Suspect minute differences in colliders.

No differences in collider geometry or Unity configuration.

Possible test cases for determining the cause of discrepancy

• Side-by-side unaltered comparison
• Replace Avionics Ring model with Service Module model
• Alter Avionics Ring model mass to match that of Service Module model

Test payload: Mk1-2 Pod + SDHI Heatshield

Side-by-side unaltered comparison (SM/AR sitting on pad)

• SM decoupler: 74->85m (11m altitude gain)
• AR decoupler: 73->85m (12m altitude gain)

Side-by-side unaltered comparison (SM/AR suspended by launch clamp)

• SM decoupler: 76->80m (4m altitude gain)
• AR decoupler: 76->80m (4m altitude gain)

Side-by-side unaltered comparison (SM sitting on pad, AR sitting on Rockomax Brand Adapter 02)

• SM decoupler: 74->85m (11m altitude gain)
• AR decoupler: 73->79m (6m altitude gain)

Side-by-side unaltered comparison (SM sitting on pad, AR sitting on Rockomax X200-16 Fuel Tank)

• SM decoupler: 74->85m (11m altitude gain)
• AR decoupler: 75->86m (11m altitude gain)

Side-by-side unaltered comparison (SM sitting on pad, AR sitting on Rockomax Jumbo-64 Fuel Tank)

• SM decoupler: 74->85m (11m altitude gain)
• AR decoupler: 80->92m (12m altitude gain)

Increasing the mass of the AR from 0.48 to 1.1 had no effect on decoupling distance.

Replacing the AR model and stack node definitions with the SM model while keeping the mass at 0.45 resulted in greater decoupling distance (74-88m, 14m altitude gain)

In conclusion:

When tested alone, there may sometimes be differences in decoupling distance, but since the Avionics Ring is typically used with a whole bunch of other stock parts to become a user-built Service Module (and becomes heavier in the process), the difference tends to be negated.

Additional feedback / info from Bomoo

• According to him, the inconsistent behaviour is not just limited to decoupling distance, but also decoupling direction when used with the KSPX LES - the command pod may jerk left or right depending on whether an SM or AR is used.
• His most recent post says he may have found the cause: Kerbal Joint Reinforcement. If this is a mod conflict edge-case, then it's probably outside my area.