I don't have an answer to your specific question, mostly because I don't believe there is an answer. The simplest application of fuzzy logic in an FSM would be allowing multiple states, rather than a single state, and using probability to determine behavior.
In this light, it's a question of whether or not an extra level of AI sophistication is appropriate. Simpler applications can suffice at times, while more sophistication in architectures provide enhanced perceived intelligence at the cost of complexity (in coding, debugging, design).
Allowing multiple states of an FSM which are executed randomly would allow for more seemingly random, yet still reasonable and "intelligent", behavior. This would have the result of seeming more human, as humans often seem random and "unpredictable".
For an individual agent, the agent might seem to do something less obvious, making it seem "foolish" or "unpredicable" or even "unexpectedly clever" depending on the situation. For groups of agents, fuzzy states might lend players to perceive the groups as "disorganized", or "unable to agree", etc.
Again, though, it's less "Which is better for individuals/groups?" and more "What level of sophistication of AI is appropriate for my game?"