The other issue is from an "event" such as wrapping a line, or switching violently from forward to reverse, etc.

The length unsupported length (cutlass bearing forward) is as much an issue as the inch-here-or there (in my opinion). The length of overhang shouldn't be proportional to the diameter of the shaft -- it should inversely proportional to the something less as I recall.

The structural analysis isn't static (though they will be modeling it that way). I think it's less an issue of "sag" and more a conditional of something torsional that is akin to buckling. As I recall it's radius of gyration (or is it the section modulus) is the key when evaluating the lenght of the shaft in combination with the position of the cutlass bearing realtive to the "point-ish" load a the prop.