It is not so much a known glitch ... more so known source of error. Sketched beams are so error prone I try to avoid at all costs. The error come from the polar moment of inertia calculation J. The polar moment of inertia J is known for simple shapes such as circles and squares. But for complex geometry this value is not known and is refered to as the "K" factor (see Roark's Formulas for Stress and Strain). It is approximated by Mechanica during the sketching phase, but this approximate can be way off. Just as an exercise look at Roarks and try to determine the torsional stiffness of an open C shape. It is brutal. If you beam is subject to any amount of torsion it can cause bad results to creep into your solution. Refer to the PTC knowlege base article for more information: http://www.ptc.com/cs/cs_24/howto/mst1359/mst1359.htm.
Secondly, Gavin also makes a point regarding where the stresses are calcuated for sketched beams. Mechanica referes to these points as recovery points. You can have up to 9 I believe. Obviously, depending on where the stress is being measured can have a huge impact on the results you seek. For standard shapes, these recovery points are shown in the simple crossection in Mechanica and they are typically located at sharp corners or on the outer most fibers of the crossection.
I fought this problem for weeks...that is how I found out about all this. My main take-away was to simply not used sketched beams if I didn't have to.