IPv6 makes the global routing table much smaller (because there's enough space for the subnets to be logical parts of the network, rather than squeezing as many addresses in as possible), so the core routers for the IPv6 internet ought to need less hardware than the IPv4 ones.

It's worth noting that during the transition period the core routers will actually need more memory and processing power, since they'll have to route both IPv6 and IPv4 prefixes for most destinations.

Ah thanks! Does this mean that the end routers can be simpler too, or do they still need significant processing power?

One big advantage of IPv6 is that routers will not have to deal with fragmentation. At higher debits, it should make a pretty important difference. http://en.wikipedia.org/wiki/IPv6_packet#Fragmentation

IPv6 also deals better with packet headers, so it's easier to calculate for routers. I suspect however this will make more of a difference for high end routers with lots of traffic.

As others have mentioned, the dual-stack transition does require more resources. Then again, you can implement NAT64 (or 464xlat), which seem to become more realistic/robust/more widespread solutions (running dual-stack also means managing two layers of routes, two layers of firewalls, etc).

(Disclaimer: this is not my day job, I just find it lots of fun to play with on my local network and the Montreal local city mesh.)

Depends what they're doing; a typical home router isn't usually CPU-bound AFAIK. IPv6 packets can be larger, but conversely that means fewer packets for the same amount of data. Organizations that embrace IPv6 can probably simplify their networking by using globally-routable addresses for all devices (coupled with appropriate security measures - NAT qua NAT was never a security feature but it did give you an obvious single point at which to place a firewall). But I fear many companies will simply reproduce the complexities of their existing IPv4 networks in IPv6, at least initially.

They won't have to do NAT. They can just route packets. Plus, the header format is much simpler, and all fragmentation should happen at the end node.

Is that really that significant a problem? Moore's Law should have dealt with memory and processing power issues severalfold in the time since IPv6 rollout was called for.

It strikes me that it's more likely a lack of strategic investment in infrastructure.

Generally speaking even now you can't do route lookups fast enough to handle provider traffic volume with current general-purpose CPUs (although it seems to be on the edge of possible). Instead routes are loaded into special ASICs (TCAM: http://en.wikipedia.org/wiki/Content-addressable_memory) that seem to not follow Moore's Law, be it due to low volume or other limitations (perhaps not technical).

Routers using TCAM seem to max out between 1-2M routes (divided across v4, v6, MPLS, etc). More recent routers are using slightly more flexible processors so there's some hope but the router product cycle is fairly lethargic.