bind only keeps track of it's own zones and any zones it's cached.
It will keep cached records based on the ZOA's published TTA.
A caching server doesn't go out and learn every zone out there. It knows the root level servers and caches everything it learns based on queries it fields from it's clients.
It doesn't check anything cached unless it expires and a client asks for that information again.
I could see it being slower for servers who use a round-robin'ing (my term) DNS server to do load balancing. I.E. the DNS server passes out different addresses for each query in some sort of even distribution.
However, even for that to work, the record has to have some sort of shelf life, or you'll get time outs during web sessions.
Unless of course if you go to
www.mysite.com and the first thing it does is redirect you to wwN.mysite.com, where N is one of N webservers that share the load.
But again, the big bandwidth is zone transfers and if you are not trusted by the ZOA, you are not going to get an entire zone transfer sent to your nameserver.
A caching server is seeded with the root level servers and remembers what it's asked for, and little if not nothing more.