I agree 10uf is large, and excessivley so when you crunch the numbers but this was required to dampen out the fluctuation i was getting from the PWM of the wideband's heater. If you take that noise out of the equation then something much smaller is required, as you have proven. I wouldnt suggest everyone uses that value just because i did.
As it happens, even with the heater noise i was getting if you had a reasonable hit count when datalogging the AFR would average out almost the same anyway - either way you are still dampening the readings whether it's done as a hardware hack or in software. The fact you need to average x readings makes a snap reading of the AFR impossible. I believe you have to step back and look at things in context, we arent tuning competition systems with massive tables and endless decimal places of accuracy though, its a relativley basic setup with the position sensors being driven via a piece of rubber & nylon.
The only true solution in my case would be to eliminate the heater noise completley, i need to spend longer tracking it down. I've noticed recently that the LM1 can be upset a lot by changing the cable length, even if you use foiled & braided mil-spec stuff for the cell and seperate supplies for the heater (although at the time i was using the stock cable). I guess there are some neat tricks going on inside the LM-1 firmware.