added another bullet on the OP:

• During the 100-minute test, with the random shutdowns I'll purposely perform, I will make sure on average the node software will be allowed to run (and I'll be doing the "E" and "C" instructions as fast as I can) 75% of the time.   In other words, 25% of the available time the nodes will be not running the node software.  While some of that 75% uptime will be re-syncing per the "E" instructions, hopefully the syncing will be quick so "C" instructions can crank out new transactions ASAP.  If the new transactions at best will reach, for example, 11.1 per second per node, and if syncing is near instantaneous (for the sake of this calculation), the best score you'd possibly get is 11.1 * 60 * 75 * 20 = ~ 1 million.