Did you refine the mesh on each model until the stresses stop increasing? You need to be sure that coarse mesh isn't causing skewed results.
Refine mesh in areas with high stress gradient until results stop changing significantly. There are options for auto mesh refinement, or you can do it manually.
When both models have reached convergence, then you will have a fair comparison between the two.
Both are incorrect. The maximum stress at an interior corner or other discontinuity converges to infinity. The stress is a function of the element size, rather than the geometry, and has no physical meaning. you can only interpret stress far from discontinuities. Define "far from" as far enough away that the stress converges to a constant value as element size decreases.
As others have pointed out, you need to make sure you're results are converging to an acceptable level of uncertainty; this is first and foremost.
Regarding shell elements, remember that a shell element is an idealized element (much like a beam or spring) and are sometimes called 2.5-D elements (I say 2.5-D because using shell elements in a 3-D environment is not the same as doing a plain-stress or plain-strain idealization). The key difference between a solid 3-D element and a shell element is that shell elements make assumptions about the stress variation across the thickness of the element. This is why shell elements are only recommended for bodies where the thickness value is much smaller than the other overall dimensions, such as sheet metal components.
3-D solid elements will always give "better" results compared to shell elements. There are cases where that difference is negligible, and in these cases shell elements will be better due to the reduced solver time required.