This is, indeed, a problem on certain types of RAC clusters. Ours is a typical example. We have a RAC cluster on HP-UX with all the shared disk on raw volume groups. If we want to write output to a flat file, we need to write it to a disk that is exclusively mounted on only one of the nodes in the cluster.
If a later job needs access to the file, this can present a problem. One way of handling this is to have the job that creates the file and all subsequent jobs that read it explicity log in to a specific instance. Then you will know that all the jobs have access to the same files. The disadvantage is that you lose some of the high availability that RAC is supposed to provide. If a node goes down jobs can fail, even if other nodes in the cluster are still up.
Another approach is to run an intermediate job after the first job finishes, which copies the file to all nodes of the cluster. Then the later jobs can run on any node and expect the file to be there.
I have heard that this issue isn't a problem if your RAC cluster supports a third party clustered file system. That would allow you to write log files and any other output to shared disk, where it will be accessible from all nodes in the cluster.