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.