Ubivis, Wronghands: I guess the current implementation is really not fit for realtime. And/or small Cell Size Side. (bigger cell side sizes over the same layout's size => less computations)
In the current state I'd say it would take another implementation of the PF to perform in real time.
I've been discussing with Yann for some time about such an other plugin/behavior.
For now, I don't really need it, and haven't really time to spend on such side project atm.
As long as you limit and delay the path generations, you may have a use for the PF (especialy in turn based/slow paced games).
Or use a "witness" object that will have the pathfinder and which pathlist the moving instances/objects of the same type (Ubivis, your yellow moving thingies for example) would look/use.
But the path would be actually generated only once on start of layout or any time the obstacles/field is modified. And not for every moving instances.
That would lighten the CPU charge for sure.
Wronghands: I saw some strange things occuring testing your exported project in firefox.
That would lead me to believe that you're not mastering the behaving of your application/logic.
Without the capx, hard to say.
And I really don't have the will/time/energy to be your personnal debugger at the moment, sorry.
You'd have more answers posting a public capx, the exported project pretty much still looks like a prototype.
If you fear for graphical/sound asset, simply make a stripped version with graphic placeholders/no sounds.