Consider cost of bootstraping. In the parallel\Runtime API the bootstrap routine is executed each time "new Runtime()" is called. So if you create e.g. 8 Runtimes, the bootstrap code will be executed 8 times and it does not depend on the number of Futures. On the other hand, in the case of Functional API each "parallel\run()" causes the boostrap code to be called again. This can be painfull if your tasks environement is complex (eg. huge datasets that need to be prepared in advance and thausends of relatively small tasks).