Let's consider a cantilever beam, which can be modeled as a solid or as a shell.
The actual deformation of the real beam (not mesh) is the sum of the shear deformation and the bending deformation as seen below:
When modeled as a solid it is recommend to have a minimum of 3 elements thickness in all places so that the shear deformation can be calculated accurately. If this is not possible because of the size of the model (there are too many degrees of freedom) we use shell elements. This is the reason we have two types of shell elements. Thick shell elements consider the shear forces (for thicker plating) and thin shell elements do not (for this sheet metal) because they are negligable and would just waste calculation time.
As a rule, thick shell elements should be used when the thickness to span ratio is greater than 5%. That's a very simple rule as long as you understand one thing... How exactly do we define span? Thickness and 5% should be perfectly clear but the span can be confusing. To explain this let's again consider the same beam. In this case the span is the length but not the width. But to be a little more clear lets consider this piece of sheet metal:
In this case we first need to assume that the verticle section is rigid compared to the horizontal section. The sheet metal may be very long and thin but because of how this load is applied, if you need to conseder the deformation of the flange you should use thick elements because in this case the span is the length of the flange not the length of the part.
It's hard to put a concise definition on span but in general I'd define span as the distance between where the load is applied and the place bending starts.