If I use a subtype, then with Lucene I have the choice of either indexing it or not, whereas if I use a type, then I can index it separately.  That is what I prefer.

More importantly, I have to do many bulk inserts and deletes -- 10K rows at a time.  Elgg's default structure of metadata and metastrings simply cannot handle that.  I need to use a more straightforward data model, with one table.

You're talking about creating a 5th entity type that has its own custom table to be joined with elgg_entities like objects, for example?  Elgg does not support this.  You're welcome to do whatever you like, ultimately, but I would try to figure out a way to do it the Elgg-y way or prepare to create a fork of Elgg.

I believe you're right.  Thanks very much for the quick help.

Time outs are not a problem so far as I know as that can be extended with the php set_time_limit() function.

A larger concern is memory leaks in core Elgg. I have found that doing more than a few hundred large profile inserts at once consumes too much memory. I know Brett was looking into this a while back but for now, I suggest that you structure your code as a cron job that does a few hundred inserts at a time. Probably 500 would be safe.

Thanks very much for the insights and suggestions!