The loading of that quantity into oracle is not going to take any less time by doing it any other way. I can appreciate a desire to speed things up, and it does seem likely that if you avoid writing that quantity of data to disk, but instead pipe it directly sqlldr, you will save the amount of time that it takes to write to and read back from the disk.

In other words, by streaming data directly into sqlldr, neither the data creation nor the database loading go any faster -- you just eliminate the intermediate delay of waiting till all the data is on disk before loading it.

That said, you might want to use a "tee" in your pipeline: have the data go to a file while it's also going directly to sqlldr, so you have something to look at and work from if anything goes awry.

You are certainly right that using DBI "insert" statements would be orders of magnitude slower.