Node Import Memory Issues - Suggestions and Best Practices

I was wrestling with giant import files (millions of records and such) and couldn't get the load to complete. I reduced the import file to 1 million records with only one field (title) per record just to see if it would complete. It still bombed after a few thousand rows.

I was just about to give up on this method of force-feeding Drupal my data when I realized that the imports were still queued up and marked as "in progress" - I got a superchub when I clicked on them and they resumed! Very sweet.

Now, with renewed vigor, I am seeking "best practices" on how to keep that barber pole spinning as long as possible without the import bombing out. Measures so far:

- I'm the only session on the box
- I jacked up my php.ini memory to like 256M or something
- I set ApacheSolr and core Search to the chillest settings and unreference my importing node type
- I whisper sweet things to the server and stroke it while it runs my batch job

Is there more I could be doing? Is it possible to leave a system unattended and have it complete hundreds of thousands of records without bombing? That would be awesome.