Thanks for your advice. I know that my problem is not disk bound as running four processes instead of one reduces the overall runtime to 25% of the serial runtime. I'm aware that forks() in principle are simpler to use if you don't have to pass information around, but in my case, I have information to pass around to the master process.

As I'm the only user with administrative privileges on the database, I will be the only one doing bulk uploads and hence I want the bulk uploads to be serialized in the fashion I proposed. Having the conversion retry at fixed (or even at random) intervals creates the risk of flooding the machine with (infinitely) retrying programs, which I dislike.