Getting conflicting results performance tuning data map pushes, any Data Management gurus out there?

Summary:

While trying to performance tune running Data maps in EPBCS we are getting conflicting results, and we are not able to zero in on an optimal batch size.

Content (required):

The test is moving data from a BSO cube to ASO cube in EPBCS.

For the same data map, we ran multiple settings for Batch size. The data being moved is very large, probably over 1Mill cells.

We ran 3 data pushes back-to-back with each batch size setting, and the variability between the timings is 2-3 minutes.

In some cases, the 5,000 and 30,000 batch size ran faster than the 300,000 and 500,000 but not consistently, we would run it once and get a low number like 8 min, then 10 min, then run it again and get 8 min. This behavior was observed for every batch size we tried.