OLEDB Cursor Questions

Howdy folks! I have some questions on cursors.

I'm trying to create a RecordSet type object in native oledb which only stores 1 row at a time. The main reason for this is that I very well may be selecting thousands of records and I don't want to run into memory issues storing all those rows. Thus I need to use a cursor to move between records in a Rowset. Unfortunately, I'm required to support SQL query processing, so unless there's an easy workaround, I'm limited to the forward-only and scrollable cursors. Both cursors would work from the functionality standpoint, but I have some concerns with both of them:

Forward-Only:

Although the forward-only read-only cursor is the fastest query processor cursor and the second fastest cursor overall, this cursor provides the least amount of functionality. It does not support bookmarks or updates.


This cursor does not allow quick restarts...in fact, it's mentioned in an msdn article that when this cursor is restarted, it may rerun the query. I would have to restart a forward only cursor to allow for MoveToPrev() and MoveToFirst() functionality. Can anyone confirm that the query is rerun on a restartposition call?

MSDN states that this cursor does not support updates, does this mean that if I execute an UPDATE query on a row within the rowset, then move to that row via GetNextRows the data will be old?

If the above is true, that makes me think the cursor is caching the data. MSDN doesn't state this, but wouldn't that be accurate?
Scrollable:

"The scrollable read-only cursor is the most functional query processor cursor. The result set of this cursor is cached. This provides fast cursor restarts but makes it more resource intensive, especially for queries with large result sets. Because the results are cached, changes to the underlying base tables for a query are not reflected in the query results unless the query is re-executed."


Is the cache the same as just storing all selected rows in memory? If so, why would I bother using this cursor at all instead of storing the rows in my own data structure?
Again if anyone knows a simple way to run SQL queries and still use a base table cursor let me know; otherwise I've got to decide between the slower two.

Thanks in advance!