When an application pool is scheduled to recycle at 06h00, if this is a 24/7 server, I can see that each instance of each related worker processes are restarting exactly at 06h00. But, when the server doesn't receive much hit or no hit during that timeframe, the recycle will only happen on the next hit. As a hit will trigger this, this is not done automatically in the background if there is no activity. So, basically, this means that the the hit triggering a recycle of the worker process will take a little bit longer. As oppose to being done at 06h00 on a 24/7 server, on the other non busy server during that time, the next hit could happen around 07h00 or 08h00, for example, and only at that time the related hit will take a little bit longer.
In order to avoid that, is there anyone of you forcing a Web hit or several hits at recycling time just so to assure that each worker process are recycling at the same time?