If you look in Monitor at the list of child pipelines executing under your parent pipeline, you will likely see that there are indeed 10 concurrent pipelines running - as one completes, a new one starts - so the start/end times of the pipelines will overlap.
The reason you see only one document difference between the input/output of the child pipeline execution is because as soon as one child completes, the next is started.
Here is a screenshot of an execution I just did that scans 24 files, then sends to a child for processing, which takes about 3 seconds per file. You can see that the last 4 files all started at 10:26:09, the previous 10 started at 10:26:06, and the first 10 had started at 10:26:02
Hopefully this helps clarify!