test.rb

require 'active_record'
require 'pg'
require 'benchmark'
require 'sequel'

ActiveRecord::Base.establish_connection("postgres://localhost/test")
DB = Sequel.connect('postgres://localhost/test')

ActiveRecord::Base.connection.execute <<-SQL
drop table t1 cascade; drop table t2 cascade; drop table t3 cascade;

create table t1 (
  id serial primary key,
  name text not null
);

create table t2 (
  name text
);

create table t3 (
  name text
);

create or replace function make_t3() returns trigger as $$ begin
  insert into t3 (name) values (new.name);
  insert into t3 (name) values (new.name);
  return new;
end $$ language plpgsql;

create trigger t3 after insert on t2 for each row execute procedure make_t3();
SQL

class T1 < ActiveRecord::Base
  self.table_name = :t1
  after_create :make_t3
  def make_t3
    T3.create!(name: self.name)
    T3.create!(name: self.name)
  end
end

class T3 < ActiveRecord::Base
  self.table_name = :t3
end

# Using ActiveRecord to handle the after create's
puts(Benchmark.measure("t1") { 5000.times { T1.create!(name: 'joe') } })
  # 5.300000   0.330000   5.630000 (  8.885517)
  

# Using triggers and no ORMs
puts(Benchmark.measure("t2") { 5000.times { DB[:t2].insert(name: 'joe') } })
  # 0.330000   0.190000   0.520000 (  0.998176)
  

# Think about how many ActiveRecord objects your tests and application code make.
# Worth the 800% overhead? Maybe not on write-heavy applications.