Hi,
dont' worry about the numer of states or actions per state. This is not in my opinion a key aspect for performances.
what really counts is what your actions are doing, that's the key. I have very big Fsm, with lost of events, lots of variables, taking to other fsm constantly, and it performs just fine.
what can really hurt your perfs are using unity features that should not be used during your game loop like finding objects, instantiating/destroying objects, bad physics set up, bad mesh and material optimizations, bad process ( whether you do it in playmaker or script doesn't matter, a bad algorythm is a bad algorythm).
Are you testing on the device or on the published version or within the editor?
It is known that performances within the editor should NOT be taken into account. Because so many editing scripts and tools are running to give feedback on what's happening, the perfs are ALWAYS poor compared to the published version. To cover this case, I have a special layout, with just the game windows, nothing else, not even the inspector, and I run there to see how it performs, this is the closest benchmark I can make before publishing.
bye,
Jean