The base client operation will produce requests based on the poll rate--4, 5, perhaps 10 per second. That covers your tag bindings and gateway=>client push messaging. Pretty much everything else will be query bindings and scripting calls that need to request data from the gateway. I would consider even 100/second to be pathological.
Audit all of your bindings to find and eliminate unnecessary polling. Where many scripts are requesting things like alarm status, consolidate to a gateway script that populates a gateway memory tag, and bind that tag in the client where needed. Find and eliminate scripted system.tag.read*() calls that can be replaced with tag bindings or indirect tag bindings.