This is fairly doable if you use a configuration management system like Chef/Ansible/Puppet - especially if you run standalone checks on the sensu-client.
This allows the clients to define their own thresholds, rather than changing the sensu servers themselves.
See https://sensuapp.org/docs/latest/reference/checks.html#standalone-checks
In this case, the definitions for the checks are sitting on the client servers and they have the choice of their thresholds or configurations. The client itself manages how often to run the check and sends the output back to the server, rather than the server requesting the checks. This helps quite a bit as far as scaling or multitenancy.
The other way to accomplish this, if you are tied to serverside checks, would be to use client attributes (https://sensuapp.org/docs/0.25/reference/checks.html#check-token-substitution)
For example, you can have a cpu check that says something like check-cpu.sh -w :::cpu_warn::: -c :::cpu_critical:::
and these come from a cpu_warn
and cpu_critical
value from the client.json on the client server.
Source: We use sensu extensively in an enterprise environment across thousands of hosts and have been working through these same issues.