One of the primary purposes of batching is to release features at a regular cadence, and all releases go through the same minimum set of checks to catch any unexpected regressions.

In my experience, the workload of these checks grows much slower than linearly with batch size.

What other factors am I missing here?