We've just uploaded a new screencast about using XtraGrid with large datasets, especially remote ones.
The screencast goes through the typical architectural approaches to downloading data to a grid, and shows how large remote datasets really kill performance due to the amount of data that's downloaded at critical times in the user's workflow. It then goes on to show how eXpress Persistent Objects, XPO, can dramatically help alleviate this issue by cooperating with the remote server to only download the records needed for display. We call this server-side XtraGrid.
For one reason and another, this video seemed to gell remarkably easily and quickly, and I'm extremely happy with the way it turned out. See what you think!