I agree with blandry.

My answer a couple of answers above, was not an answer.

It was kind of trying to find a solution so that your production servers are sizeable.

Saying that, we can manage to develop a load test with x millions of rows that would be consider a full-size production server.

Once we are near to reach this amount when on production, an alert would tell us to set up several servers to load balance the loads.

So although theoritically, you can never know the amount of data you will deal with, you can "manage" it by spreading the loads on different servers.

By doing so, the load testing makes entire significance!

The sizing of the load test then depends on how frequently you will extract, purge, reallocate ressource.