Keeping per process CPU utilization in memory of the statistics collector should actually allow to identify whether a process consumes significant CPU time.

This request may sound similar to ZBXNEXT-88, ZBXNEXT-164 and ZBXNEXT-494.
But this it is primarily not about getting the top <n> busiest processes nor the aggregated utilization of a defined process (group).

The aim is to identify the one or several processes that by itself resp. by them selves consume more than <n> percent CPU time over a period.

Maybe something similar to this:
proc.num[<name>,<user>,<state>,<cmdline>,<cpumin>]
cpumin - each process consumes more than <limit> percent CPU (core) time for one minute.

Of course it could also become an enabler of ZBXNEXT-88, ZBXNEXT-164:
proc.cpu.limit[<limit>,<format>]
format - pid (return list of process IDs), name (return list of process names), full (cputime pid user cmdline)