I don't have time to comment just now but the comparison between the different fetch methods and data set size isn't all that appropriate for PostgreSQL. There you'd use ORDER BY/OFFSET/LIMIT instead of the $max_rows. You'd also contrive to use a cursor which means you'd probably eschew DBI and use Pg or Pg::Simple. That is, if you found yourself working with really large result sets.
The DBD::Pg driver always has PostgreSQL complete the entire SQL command and then copies the entire result over to perl. Trying to iterate with ->fetch doesn't help with that - you'd need a cursor (which isn't supported under DBI) or to reduce the data size.
Anyone else care to follow up with some concrete advice? I've got to head off to work just now.