Hello Brandon,
it depends from:
the real WAN speed circuit
the router platform you use
The first point is clear: on a STM-1 155 Mbps you could use a 95% max-reserved-bandwidth on a 64kbps things are different.
About the router platform: some high end router platforms (like old 7500 but GSRs too) actually miss this concept of max-reserved-bandwidth in this case you need to configure your QoS scheduler by using a class for control traffic including eigrp and you can choice to allocate a 10% bandwidth.
This can be an approach used also on platforms that actually implement the max-reserved-bandwith that have a hidden system queue for routing and control traffic.
If the network is stable EIGRP bandwidth consumption is very low and limited to hello packets.
If in the network there are some instabilities the EIGRP diffusive algorithm will start to send Queries and expect timely answers from each neighbor.
You can limit the query range by using the EIGRP stub router concept and appropriate route summarizations.
If your only routing protocol is EIGRP and the link is >= T1 speed I would use 90% as max-reserved-bandwidth or I would use 100% with a class for routing control packets with bandwidth 10%.
The advantage of the second approach is the increased capacity to monitor your QoS behaviour with sh policy-map interface you can see statistics for each traffic class including drops so it provides you a feedback that helps to understand if your tuning is correct. The difficulty is to list and classify all the control traffic in the added class map.
Hope to help
Giuseppe
