have you considered mesh size?
are you using static or nonlinear analysis?
you may want to post your model or some pictures of your models and then show some of your back end calculations that show what you're comparing against.
overall, i could potentially see some high stresses happening in the area of a shrink fit just because of how the software handles them. but similar to other high stress areas, you may need to interpret the results a few elements away to understand exactly what is happening. some research on singularities and mesh convergence would be good. here's an article we wrote awhile ago:
You would have to run a non-linear material to get a better feeling for what is going to happen. The material has to yield and flow to form around the insert.
You are going to have very small radii (probably 0 in your model, if not in the real parts) at the serrations.
Knit lines are often a reason why some bosses fail and some don't. Bosses which have weak knit lines will fail much more often than bosses with strong knit lines. Processing differences can cause knit lines in a given location to be much weaker than normal.
Cracking can also depend on how the inserts are inserted. Are you doing it with heat or with ultrasonics? Using too low a temperature or too low ultrasonic energy can cause cracking.
The "right" way to run the analysis would be to run a multi-physics analysis combining thermal and mechanical analyses. Even then, how would you account for knit lines?