Have a look at the different conditions which drive the different CPT's, that may give you some clues.
Take the example of a tank level PID controller that's set to keep a tank level at 50% by opening an inlet valve more or less. Now let's say that whatever liquid is in that tank will at some point be pumped out very quickly to "flood" some kind of process, but several minutes later will be returned to the tank. What you don't want is for the PID to ramp right up to 100% trying frantically to get the level back up to 50%, because you know that (a) in 2-3 minutes, all that liquid is going to come flooding right back in, and (b) when it does, you don't want your PID loop to be hanging around way up at 100% and having to slowly ramp down, when the level is already way over its 50% setpoint. So, during the "flood and return" part of the process sequence, you limit the output to 30% - enough to still trickle in and ensure the tank doesn't run dry, but not so much that it's going to end up oscillating and chasing itself for 20 minutes after the "flood and return" process is all done.
This is perhaps a slightly over-simplified example, but hopefully it illustrates the idea. I have used limits in this type of scenario on a slightly more complex temperature control loop which had a very strict tolerance on overtemperature conditions (e.g. overshoots of more than 1°C were undesirable, and had to be limited to less than one minute). At known times, there were sudden and significant changes in the airflow, which would very rapidly affect the measured temperature, so I implemented limiting to ensure the PID loop could respond quickly enough to the significantly changed process conditions, while still leaving it tuned generally non-aggressively for better temperature control at normal times.