Some results of the test gleaned from
this plot:
Estimated network capacity: 85 kB/min
Typical input tx rate before test: 30--50 kB/min
Typical queue size before test: 300--600 kB
Peak input tx rate during test: 126 kB/min (~3x normal, ~50% over capacity)
Peak queue sizes during peak test: 12 MB (14:00), 14 MB (21:30)
Sustained input tx rate for several hours after peak: 70-100 kB/min
The test definitely showed that even a small player can create a large backlog with modest expense.
It showed that when the input transaction rate is close to the network capacity, even a small increase in that rate can create a huge backlog. Between 18:00 and 21:30, when the input rate increased from ~75 kB/min to ~112 kB/min (a 50% increase), the queue grew from ~3 MB to ~14 MB (a 370% increase).
The previous stress test (on a late friday night) used only free transactions, so the fee-paying transactions were delayed only slightly. This one used fee-paying transactions: it will be interesting to see how it affected the ordinary fee-paying transactions. I hope that someone will do a detailed analysis.