Multiple llListens VS monolitic

Here's a question:

Suppose that I need to run a script that listens to multiple selected people out of a large crowd (say, 20+ specific people out of 40). Which design of the following will cause the worse Sim impact?

A global llListen(0,"","",""; with an IF statement checking for keys on listen events, or a master script with several optimized child-scripts, each running their own llListen(0,"",key,""; filter that listens only to that specific person?

Remember that the crowd is chatty and thus there's expected to be alot of listen() event triggers. I am trying to find the best way to minimize the impact for the event duration.

My theory regarding the second part is that if I can close the listeners whenever the need to listen to specific people ends, then the sim impact can grow bigger or smaller depending on the need - where a monolithinc global listen would continue to trigger on every little sentence...

... or could it be that running a whole bunch of silent llListen filters on separate scripts is worse?

Whats your take on it?

Can you filter on a non-0 channel instead?

I don't think that changes the impact in any way. A listen impacts a sim, regardless of people talking or not, whatever channel they're on.

That said, the monolithic approach would be FAR more server friendly.

A single listen with perhaps a llListFindList call probably would be better than a giant 20 deep if/else bit.

I am infact allready using llListFindList in the monolitic llListen - but I am pondering if using a bunch of llListens with a filter in their field is any more efficent then a single llListen that is completly open on channel zero.

Hiro, ofcourse I could, but that does not answer the question:
Which is more impacting on a Sim, in the specific event of having to listen to only several specific people out of an entire room filled with chatting people: A single unfiltered Listen with a scripted selection inside the event itself, or several independed llListens with the filter on their command parameter?

It has been found that a single listen with the logic performed manually in the listen event is more efficient.

I don't know what kind of an impact it would be to the users themselves, but I'm thinking about a similar need for this type of thing. I'm planning to do the monolitic approach, but will use a random non-zero channel for the people to interface on. That will not change the amount of llListen lag, but internal to the script, the "if then" code will have less to filter out. Since many text operations seem to also slow down a script, sorry I have no imperical or statistical measurements for this, I think this will be at least marginally better.

Well, for that matter, any script "impacts" a sim. It's a matter of how much. A listen with a filter affects a sim less than one without a filter.

However, two listens with filters impact a sim more than one without a filter.