I think I have a mistake in here (at least for part of it) I think the table log is misleading as I have it set to 100 rows so I may have lost a few seconds off the top due to it exceeding 1000 rows.
However this doesn't explain the other table where I appear to have failures coming back, but no requests running.
I've tried this test again this morning and I don't get this behaviour. I get an increase in the number running, before the failures start to come back, and I think the stats section is consistent with this.
So maybe this is just spurious result. Would be interesting to see your comments on the unexplained bit though.