If you have multiple workers writing to the same file and you have not taken care to avoid them overlapping in the file, then you could get file corruption. In particular, do not have multiple workers save() to the same file without going through an interlock to ensure that only one worker is writing at a time.
If you do need multiple workers writing to the same file, then sometimes the easiest approach is to use spmd instead of parfor, as spmd can use labBarrier() to synchronize workers. On the other hand, with spmd is its often easier to designate one worker as the worker responsible for the I/O to the common file, with the other workers using labSend() to send it the data to write, and receiving any read results using labReceive()
With parfor and fairly recent releases, a way to avoid having multiple workers write to the same file is to use a data queue or pollable data queue to send results back to the client which becomes responsible for doing the writing.
If the workers need temporary files but the files do not need to overlap between the workers, then the easiest way is tempname() to get a unique file name. If you need a directory of files but they do not need to overlap between workers then use tempname() to get a unique name, and then mkdir() that name, after which you can write into that directory.
