After the upgrade, did you rerun a statistics collection? It's a bit hard to say how much of each index is going to be cached. That depends on the workload and also the query execution plans.
An additional thing to try is to set the optimizer back to 11.2 and see if the behavior goes back.
e.g. (the following assumes your old DB was 18.104.22.168)
alter session set optimizer_features_enable='22.214.171.124';
Not sure at all to understand what do you mean... Hence, I have more questions than answers ;-) Sorry about that.
How did you check that "PCSGM index being used more"? More executions plans are using it? More LIO? More PIO?
Which metric do you use to check "the index size fluctuating a lot"? In other words, what's the y-axis of your Chart?
About caching... How is your buffer cache sized? If it's automatically done by the database engine there are chances that with the upgrade from 11g to 12c less memory is available for the buffer cache...