I, too, have mulled over this knotty question, doing a fair bit of inconclusive stress analysis. Several years ago, I decided that the only way to resolve it was to build 15 to 20 webbed specimen spars and test them to destruction. The test spars had hard balsa flanges (top and bottom longitudinal members), with various designs of balsa webs. Length was about 20 times depth.

The commonly used balsa shear webs with grain oriented vertically failed miserably in the webs, at about 25% of the calculated load. (Very distressing, but model design surely keeps one humble, if nothing else.) The load was applied in the middle of the spar, and each end supported, since the ideal uniformly distributed load would be much trickier to achieve experimentally.

Web failure always initiated near the ends of the spars, where the shear deflection was greatest, proceeding inward. With web grain at 45 degrees, the webs were about twice as strong for the same weight. The compression flanges of the spars were restrained from buckling by a continuous slot into which the flange was fitted.

Best arrangement proved to be ply balsa webs, with grain direction crisscrossed, as in normal plywood, and 45 degrees to the longitudinal axis of the spar. For the same weight, this was about four times as strong as single ply balsa webs with grain oriented vertically, but I will have to consult my notes for more definitive figures