I would think it depends more on the way you are accessing the data than the
size of the database

so if you are selecting or updating 10 rows, this will be fast if you are
using an index and slow if you are not using an index, regardless of which
method you use.

if you want to actually retrieve a large volume of data into the application
then, as you suggest, using the lowest level primitives with the least
overhead will give the best performance.

of course for an application like myspace with a large number of concurrent
users, the data access technology will have a significant impact on
performance, but this is more to do with the level of concurrent activity
than the size of the database.

Andy