The puppet splay doesn't really work for us; instead, we're restarting puppet once per day, 
but at a random time that depends on the client's fqdn and uniqueid:

$hour = fqdn_rand(24)
$minute = fqdn_rand(60)
$second = uniqueid_rand(60)

       cron { "puppet":
               environment => "MAILTO=null_user",
               ensure => present,
               command => "sleep $second && /etc/rc.d/init.d/puppet restart",
               hour => $hour,
               minute => $minute }

so this basically does the same thing as splay, spreading out our ~900 clients so they don't 
all hammer the puppetmaster around the same time. It also ensures that they are all running 
-- we had problems early on with the clients dying because of OOM