There are 4 power modes in which an RDRAM device (aka just the RDRAM chip, not the entire RIMM) is in.
Sleep, Nap, Standby, and Active. When not much is going on, ie the device has already sent off the required data, it leaves Active mode, and goes into sleep. The next time the device is accessed, it moves from sleep, to nap, to standby just before the data is sent, and then to Active when it is sending.
Obviously to move from each mode takes a little bit of time, and adds to the access latency. If you switch RDRAM to Nap mode (which uses more power and creates more heat than sleep), you reduce the latency and increase the performance of a read or write operation to RAM, as each device only has to change modes twice instead of 3 times.
The trade off is heat and power consumption for better performance, but with the heat spreaders and massive power supplies that everyone is using in their RDRAM systems these days, this is not really a factor.