Well, it depends on what you're extending. For example, RDBO has a very flexible framework for defining new column types, encapsulating inflate/deflate, custom DBI bind_param()s, type-specific attributes, and custom SQL for select, query, and insert into a single logical unit. DBIx::Class is still working on some of these features, and the lack of a standardized way to add support for new column types has led to less support for "rich" column type handling.
As in any large collection of code, there are really only a few performance-critical parts of RDBO that are aggressively optimized. The vast majority of the code is quite modular.