The point of using 1.0 fs (instead of 2.0 fs or, for that matter, 0.5 fs) is simply to be consistent with the simulation protocol MacKerrel and collaborators have used to test the properties of their force fields. Other than to simulate fast X-H vibrations (which would otherwise not be numerically stable), I don't see any good reason to use a 0.5 fs time step. The force field will not get better if you decrease the time step.
I was mainly speaking about the accuracy of the integration itself, which is known to be timestep dependent. Naturally, the timescale of the motions being sampled has a bearing on this.
The force field development strategy is certainly another good reason to use 1 fs, but that choice in turn was based on the accuracy of the integration. I agree that there's usually not much benefit to using 0.5 fs, and probably none at all if SHAKE was used to constrain X-H bonds.
(I also don't think that all of FF development was done at 1 fs, although most of the more recent additions have been.)