Speedup from parallelization is limited by the sequential portion. If % of work is sequential, maximum speedup is x no matter how many processors you add.
Formula: Speedup = where S is sequential fraction, N is processors.
Implication: identify and optimize the sequential bottleneck first. Parallelizing the rest has diminishing returns.