The simulation has obviously too few points. This doesn't necessarily explain the deviations from expected filter curve, but higher frequency resultion is necessary to understand the simulation results.
I'm not a FEM expert but would expect too coarse meshing or errors in the structure definition causing the deviations.
Secondly, how do you know that the specified filter characteristic is feasible as 4th order? Stop band attenuation isn't included in the data entry and as far as I see from a general filter tool would require 6th to 8th order for the given ripple and pass band.