If you get the records in chunks, it's really simple to write a sipping pipeline to take each chunk and write it to a file. You close it when you're done with the chunk. And you can decide whether you should append or rewrite the file. That's also helpful when you have some header record rather than a key in each data record. As John points out, when you're writing to SFS make sure to say so rather than using a file mode.
The alternative is to use ADDPIPE to create a separate pipeline to write to a single file; you cascade the selection stages to pick the ones this pipeline needs to handle. That's attractive when records for different files are interleaved and you want to rewrite the file, but can't tell in advance which ones to wipe. I would probably opt for this approach when the chunks are pretty small and the number of files is not too much. Another interesting twist is using DISKUPDATE when you can produce the record number; when dealing with RECFM F files you can even write in non-consecutive order or create holes in the file. Sometimes it's easier to just create what you need, rather than figure out whether someone else wanted just what you need. You normally learn a few more tricks while doing that, which helps with future plumbing challenges. Sir Rob the Plumber