Brodie,
A couple of things:

The CAT 3 is for the MCC as a whole, which is meaning the bus. In the bucket you have small wire leads that come off the stabs, and that short length will limit your fault current more than you realize. It would not take much to calc the IE on the load side of that wire, even if you just model it for the largest one, assuming no more than size 3 or 4.

In a combination starter that employees a breaker, that breaker usually only has an 'instantaneous' function to prevent false trips. The motor 'overloads' (also called heaters) provide the long time trip functions. Table 430.52 of the NEC sets the max limit for the size of the circuit protector versus the motor FLA. For instantaneous trip only breakers it is 800%. If its more than that then its not code compliant.

As far as the AIC of the breaker. You said they were 'overdutied' not sure what that means. They have to have a minimum AIC that will handle the fault current at that point. Again, the fault current 'at that point' is going to be less than the fault current at the bus. The breaker has to have some AIC rating you just have to find it or look it up. So tell the owner that there is a chance he may be able to solve two problems by doing the bucket calc, which may lower the IE for live work on an open bucket, and make his existing breakers now code compliant. Maybe!

Cat 3 on a typical industrial MCC, other than something huge, would imply to me that your overcurrent device has too much time delay. You are operating in the short time band of the trip. There are only limited reasons, like small MCC lineups with big big across the line starters - that you can't employ an Inst trip set on your MCC. I was running 800A feeders to a MCC lineup, that had Inst trip, and was starting a 300HP motor on a wye-delta starter. It never came close to drawing the 6500 amps needed to hit the Inst trip point.

A lot of times the company doing the analysis does just that - the analysis. The numbers fall where they may. A good analysis will also point out opportunities to reduce the IE. I've taken MCC's with a Cat 3 to a Cat 1, by changing a $1200 programmer on a switchgear breaker to one that only had 'short time' to one that had 'short time' plus Instanteous. That was two years ago and I talk to the electrician regularly - they've never tripped the breaker. And I knew they wouldn't. So there is lots that can be done.

As for the policy, written versus what you are instructed verbally, a hardbound log book that is used regularly and consistently can be submitted as evidence. So if you are placed in a situation that the owner commands you to violate his own policy - I would snap to attention, say "yes sir", then write down the request, date, name, reason, etc in my log book, don my FR, and do the work. You show up with your log book with 50 entries after someone is hurt, and you'd be glad you had it.

There is a good chance that the 'infeasible' definition may be applied to this company's MCCs. They just need to perform the analysis. There is nothing like having 3 or 5 engineers, ESHa, production people in a meeting, and have the notes of the meeting documented and added to the appendix of the Arc Flash Analysis Handbook to show why the conclusion was made. A lot of this is about following procedure and conforming to the intent of the code.

My 2 cents.