You omitted stray inductance on the signal and ground lines between parts. In essence you've made a stack where nine capacitors are phased inside of each other somehow.

Figure ballpark 1nH/mm equivalent for trace length. If you model the ground network as well, you can predict common mode noise to some extent.
Paralleling different values is a commonly repeated trick but it's almost never productive (as you can see from the ~0.7 ohm mid range peak here).
Much easier design -- put one relatively beefy ceramic as close to the rectifier and switch as possible, put the film in parallel with it for bulk, and add a series inductor to another ceramic capacitor. Finally, bring around the power input pin (assuming this is a non-isolated supply) and LC filter it to the same common ground point. These three connections, isolated from switching noise with inductors and coupled together with capacitors, approximate a point. A point, having zero dimension, also has zero voltage drop across it, so in this way you eliminate common and differential mode noises (to within whatever attenuation curve your filters exhibit).
Tim