I suspect that I was a fool when I tested that method, and that the sort_id is a perfectly reasonable approach. The dynamic paging text is from something I wrote awhile, and I didn't retest my conclusions - my apologies.

I'll correct the benchmark data provided and add a note about your correction to the end - thanks for catching my error.

Update - Redid benchmark

The sort_id approach works better. Thanks again for pointing out the error.