Sounds like a tunning issue to me. The probe is in the correct spot then. If it was in the mash you introduce a lot of lag to the response time, hence jmer's method.
A forward feed from the mash tun would allow you to reach your temp a little faster but, I haven't seen any economy priced PID controllers that support that.
If you're fluctuating 2 deg F at the RIMS tube by the time it reaches the mash it will average out rather quickly.
Oh, I bet I know what you're problem is! There is a time variable that you need to set some times. This is how long the output should stay on at a minimum. I think Auber PIDs default to 2 seconds. If you set that variable to 0 it goes down to 0.5 seconds minimum on time. This is for things like compressors so you dont short cycle the hell out of them. If you have a different PID read the manual and look for a minimum on time variable.
What happens is the overshoots but cant turn the element off because it hasnt been on long enough.