Replication

Hitting the theoretical limits will probably be hard. Your hardware will be more of a determinant. The usual suspects of RAM, storage I/O and network will probably be maxed out before any documented limits are hit.

If you have very inactive publishers you will be able to support far more than if you have extremely high transactions just as LAN replication will afford more load than that running over a WAN.

EDIT

Remember that constantly starting and stopping uses additional CPU/memory over running continuously.

The bottom line is that every environment is different and there are no hard and fast rules to help here except: baseline, monitor and adjust as necessary.