Consider the following scenario: An initially serial application, with a larger than L$2$ but smaller than L$3$ working set, is parallelized across N cores. Of the serial application time T$,$ P seconds are parallelizable. Is a parallel latency of $($non$-$trivially$)$ less than $($T$-$P$)+$ P$/$N possible, and, if so$,$ how$/$why$?$ If not, why not?