There's your problem. Experimental rilkef isn't suitable for the operation of nophertrunions, no matter what the benchmarks suggest. In my experience, the speed of experimental rilkef is unparalleled, but there's too much side fumbling to use in a production environment.

The units on the graph are misleading you. Notice that both measurements were multiples of 100. They were probably just measuring the load on the prefabulated amulite.

The graphics are misleading in that both values are actually the same but we ran out of toner, so later we changed the numbers to fit the partial graphic. As long as the visuals agree, the datumses are unimportant.