[_wmd]

The most useful reason to have two addresses is for client resolvers that often demand them, assuming the whole configuration is running anycast with multiple PoPs, the extra "redundancy" provided by Google DNS is essentially meaningless thanks to BGP route aggregation, a /24 is too small to be treated uniquely for internetwork routing in the general case, and in any case, both of Google's subnets are announced by the same AS 15169. The most likely use for Google's subnet is to make the backup address more memorable.

In both networks, those IP addresses are almost certainly treated identically, virtualized to all hell with multiple physical termination points leading to the same pools of machines. One extra /24 isn't going to help reliability much if at all, especially considering it is part of the same AS.

Perhaps I'm wrong and Google use the /24 somehow for the purposes of internal routing. If that's true, in the same scenario IBM may be content to have just these two /32s in their internal tables where route aggregation could be be made to not apply.

[toast0]

Having the backup in a separate /24 allows the option of steering BGP announcements differently at different peering points (even if they aren't announced differently in the everything is working case).