Good point. There is a trick to avoid a divergence.
Actually the radius of the cylindrical wave is set to the spot size at the surface of the crystal instead of an actual beam waist. This is the idea Dmass told me before.
So that the radius is expressed by w=w0(1+(L/2ZR)2)1/2, where w0 is beam waist, L is the length of the crystal and ZR is the rayleigh range.
In this case the radius can't go smaller than w0/2 and the solution can not diverge to infinity.
Why does the small spot size for the case (A) have small efficiency as the others? I thought the efficiency goes diverged to infinity as the radius of the cylinder gets smaller.