Partially not kidding - I've used this as far back as 2004 to build cron jobs that pull a dataset, split it into N subsets, then fork out into N children to process those subsets of data in parallel. It's certainly not fluid like tasks and it creates a lot of memory duplication if you fork AFTER your data is pulled, but it works surprisingly well. Parent thread can wait for children to finish and even message children through socket pairs (or pcntl_signal for very basic stuff). I agree that it would be nice to have a native abstraction for tasks with messaging and concurrency limits built in.