I have done a few deployments for different clients, and I have not seen any adverse impact to the servers performing collections. That includes client access, edge transport, hub transport, and mailbox servers.
That being said, every environment is different, and I always recommend testing on a candidate server (preferably in a model environment) to make sure it plays nice with your servers & resident applicaitons and doesn't introduce adverse load (which I am yet to see, but due dilligence is always a good idea).
From there you can extrapolate how much data you should expect to see in a production deployment & plan accordingly.
I hope this is of some use to you 🙂
