Problem with a new role

Hello.

We've been using capistrano sucessfully for a while now with the basic
roles (:app, :web and :db). Now we added a new server with a custom
role :import, that should have the actual release on it, but no
mongrels running.

The roles are configured like this:

role :app, 
"root-Q0ErXNX1RuYjwlwYZV3FlQ@xxxxxxxxxxxxxxxx","root-Q0ErXNX1Rub7JcoBIUF6mw@xxxxxxxxxxxxxxxx",
"root-Q0ErXNX1RuZrj7GURp3nyw@xxxxxxxxxxxxxxxx"
role :web,
"root-Q0ErXNX1RuYjwlwYZV3FlQ@xxxxxxxxxxxxxxxx","root-Q0ErXNX1Rub7JcoBIUF6mw@xxxxxxxxxxxxxxxx","root-Q0ErXNX1RuZrj7GURp3nyw@xxxxxxxxxxxxxxxx"
role :db,  "root-Q0ErXNX1Rub7JcoBIUF6mw@xxxxxxxxxxxxxxxx", :primary => true
role :import, "root-eEed58MKCH+l5wQoFSNtmw@xxxxxxxxxxxxxxxx"

The relevant tasks for handling the mongrel clusters are:

namespace :deploy do
  task :start, :roles => :app do
    mongrel.cluster.start
  end

  task :stop, :roles => :app do
    mongrel.cluster.stop
  end

  task :restart, :roles => :app do
    mongrel.cluster.restart
  end
end

namespace :mongrel do
   namespace :cluster do
     [ :stop, :start, :restart ].each do |t|
       desc "#{t.to_s.capitalize} the mongrel appserver"
       task t, :roles => :app do
         if stage.to_s == 'staging'
           run "mongrel_rails cluster::#{t.to_s} -C #{mongrel_conf}"
         else
           run "monit -g mongrel #{t.to_s} all"
         end
       end
     end
  end
end

But when i deploy just to the import server with 'cap ROLES="import"
production deploy', capistrano always tries to restart the mongrel on
pimp01, even though it is definitely not an :app-server.

Why does capistrano think it has to restart the mongrel cluster on
that machine, and how does one configure this scenario so there's no
mongrel handling on pimp01 at all?