Redundant taxonomy enumeration causes significant performance degradation.
In used_organization_ids and used_location_ids a call to get_taxonomy_ids is used with which_organization/which_location sent in as a parameter respectively. It looks like which_location already contains all sub-locations and needs not to be expanded further. Each expansion like this causes another sql query per each sub-location.
How to reproduce:
Create a big locations tree with 3 root nodes:
Assign around 100 subnets to a single sub-loc (for example LA)
Now add only LA to a non-admin user.
Perform api/v2/subnets query with this user.
Clone the user, but add to it all three root locations (America, Europe, Asia).
Perform api/v2/subnets query with the cloned user.
Observe time difference between requests