IMO, though both servers may be receiving same number of events/operations, all these events will come to primary Hub on S1. All the porcessing of these events will be done on S1 only. Like adding events to the gateway queue, dispatching the batch to the listener by peeking events from queue. And possibly this listener is also processing the events dispatched to it.
Compare to S1, S2 is not doign lot of work for gateway events. So it is for sure that S1 will be utilizing more CPU than S2.
But as JDeppe said, 80-90% of CPU utilization looks much more than expected. Is it like listener is doing CPU intensive work?
Which gemfire version you are using?
Does server have partitioned Region?
From 7.0, WAN module supports parallel wan where each nodes dispatches batches independently. In 7.0, Parallel wan is supported for partitioned Region. If you use parallel wan then both S1 and S2 will dispatch batch to their listeners independently.