Because soil does not have a linear poisons ratio. But there are these "averaged", experimentally derived values which appear to work for most cases. As well as being appropriate for numerical solving (going close to 0.5 is a problem, 0.2 is much more stable).

You can always do a triaxial test to calibrate your poison value (preferably using an "improved" or "the Danish" triaxial apparatus). However, be prepared to see poison ratio go beyond 0.5 during dilation, and below 0 during some random spikes of data, during unloading / reloading. Luckily, strength is not affected much by poisons ratio.

Loads of thanks for your answer. Could you elaborate a bit more on the concept behind your notion hinging around being a problem and being in a stable condition when Possion's ratio equals 0.5 and 0.2, respectively? 

So, the original limitation is that going above 0.5 will force your FEM elements to "explode". They will be expanding while being compressed, which will cause numerical instability. Thus plasticity is required to produce dilation. And it does not matter how low the poison ratio goes - plastic component can compensate for it.

For drained soil poison ratio of 0.2 just happens to produce a "similar" stress / deformation field as is experimentally observed. And the dilation is projected on top with the plastic flow rule, which takes care of the plastic increments filling in the otherwise missing dilation and plastic response.

In comparison, undrained soil volume is near constant, thus measured poisons ratio is 0.5, but the solver has problems with 0.5 as well.  Thus, we usually use values of 0.499, being close to 0.5 helps simulate a realistic "elastic increments" for the stress field.

General remarks:

Be aware most sand properties change over time and HS was never intended to capture full scope of cyclic soil properties. It is compatible only with "small" unloading / reloading. In our lab, we are investigating soil stiffness "recovery" during large, drained loading cycles. This is done to develop less conservative design processes, and to improve upon models such as HS model.

HS is meant to capture very little - bearing capacity, and (arguably) improved hardening curve. Unloading / reloading is grossly oversimplified, volumetric response as well. Real stiffness and volumetric response are not linear or constant, and poisons ratio is not a property of real soil at all - it is a parameter required for "guessing the preliminary stress field" to derive the next plastic increment from.

With that in mind - tests run with poisons ratios of 0.2 and 0.3 will give similar results. It is often generalized, that compressible soils should have lower poison ratio, for no other reason than being compressible. But here I attach a figure of one of our recent triaxial tests, where long cyclic loading was done and volume response was recorded (void ratio, to axial strain plot). It shows the hidden complexity of real soil volume response during cyclic loading. HS can not and was never meant to capture it.