High response time on one Pod


currently we have some problems with high response times and some 503 codes on one pod.

Here our application is running four identical pods, somehow some nodes are faster then others.
The blue pod here is the slowest one, the node is using 100% of its cpu.

Is there a way to improve the load balancer so the yellow and green pod will receive more RPS then the others?

Is the LEAST_REQUEST load balancer the foreseen solutions for this problem? Or do i understand this wrong?

Is the request duration that high because of many connections in the waiting queue?
Are there any metrics i can check?