That is what I thought at first, however it’s not as simple.
Imagine comparing a substrate-gas system to a substrate-fluid system when I compare the two results I get on one hand 2surf_tens_SV + restraints_contribution, while on the other hand I get 2surf_tens_SL + restraints_contribution. Even assuming that the contribution of position restraints is the same in both cases, the resulting system is under-determined.
What I did initally was to compare a solid-gas (‘vacuum’ really) configuration with a soild-fluid-vapour one and that also lead to an underdetermined system (even provided one has an estimate of the equilibrium contact angle between the three phases). Doing a pure solid-liquid NVT simulation is not so trivial on the other hand because one would need to make sure liquid density exacly matches the one of the film in the S-L-V configuration.
Anyway, I believe these are now only technical issue; the most fundamental problem was if the surface tension calculation itself made sense when using position restraints, which now I believe it does. Thanks.