For example. I have a lot of CISCO routers with OSPF protocol. Thnis crazy IOS use highest loopback interface address as router-ID address; I use loopbacks to install load balancing etc. and I can't prevent loopbacks from being equal on the different routers. That's why I hardly need some IP addresses for 'Loopback 98' interface to use it as router-ID; and this have to be higher than any user's addresses. I use 233.255.254.0/24 for this purposes, but it's not reserved address.
This is one, simple, example why it's nessesary to reserve some short address space in the begin and in the end of total addresses.
No, that's an example of a poorly designed protocol implementation. One ought to be able to specify an arbitrary router id for OSPF (heh - even Bay routers can do that :) rather that relying on such an odd algorithm. I was so surprised by this that I just had to go look it up: I know _it's example of poorly designet software_. But I'd like to note it's not only example when it's usefull to have some addresses _greater than any other_ for private usage.
<http://www.cisco.com/univercd/data/doc/software/11_2/cnp1/5ciprout.htm#REF38888>
The equivalent Bay reference:
<http://support.baynetworks.com/Library/tpubs/content/114065A/J_55.HTM#HEADING55-6>
Yes, I was more surprised when they (cisco) did not implement something like _ip ospf router-id A.B.C.D_ into new IOS 11.2. We have 3 or 4 routing troubles due to this IOS property (and it always looked as _hidden bug_ because it is si,ular to the delayed bomb - it explodes 1 week below some mistake was made in the config files -:)).