I think the reason that we use an averaged TSF is that in general we assume that honeycomb panels will have biaxial loads and the buckling calculations don't distinguish between Nx and Ny loads, they simply use the biaxial load state when calculating the buckling critical load.

I suppose what we could do is calculate TSF in the x direction and TSF in the y direction and then weight them by the magnitude of compressive load.

Something like:

TSF_used = Nx / (Nx+Ny) * TSFx + Ny / (Nx+Ny) * TSFy

So that way for pure Nx loads, TSF = TSFx and for pure Ny loads, TSF= TSFy

I could see implementing something like this, what do you think?