It would be interested to know which threads are running that are
consuming so much cpu (onstat -g act) and what are they doing (onstat
-g stk <tid>).

You could try some profiling of the oninit as well to gather more info.

On the other hand maybe this cpu usage is not a bad sign. As long as
when the real processing begins, whatever these threads are doing they
don't interfere with the user threads and are scaled down to let the
sqlexecs threads do their work.