When we started with splay, over time we found that puppet runs would flock together. If there were network or system load issues causing multiple puppet runs to be slow, they would seem to clear at the same time, then be on the same schedule from then on. As other clients would hit a slow run, they would eventually find themselves in the flock with the others. The foreman chart of number of runs over past few minutes showed this clearly. Our solution was to cron a weekly puppet agent restart on all clients, then they re-splay and life is good.
When we put puppet in crontab, we had puppet manage that and used fqnd-rand to randomize the minute value.