Max traffic inbound bind cap?. Absolutley bizzarie.

[Two Dog]

Hello there, we currently have 2 recursive resolvers running SOL9 and
Bind 9.3.0 that is preplexing me to no end and I'm sincerly hoping
someone can point me in the right direction as I'm near my wits end. 
Our customer state theyocassionaly has to reload the web page to have
it resolve properly.

The named.conf is pretty basic.

options {
        directory "/opt/apps/bind-9.3.0/";
        pid-file "/var/adm/named930.pid";
        statistics-file "/var/adm/named.stats"; 
        dump-file "/var/adm/named_dump.db";
        minimal-responses yes;
        recursive-clients 2000; 
        notify no;
};


These machines are customer facing and see about 4 million queries an
hour or about 70,000 a min.  The majority of these about 45,000 being
A record lookups.

My MRTG graph is showing that the upper limit on inbound traffic to
these servers is 2.11 M a second and at times here's the weird part
it's sporadic and inconsistent that it can tqake as long as 2 to 5
seconds to resolve a lookup.  Is there a limit on the inbound traffic
that bind accepts?  What would be capping this?  I'm pretty perplexed
about this because it simply doesn't make sense to me.  I don't
believe it to be a network issue, the interface are full 100.

I would really appreciate any thought or pointers, even constructive
critisim at this point that would lead me to a solution.  Is it
possible that bind is maxing out it's open sockets that it accepts
traffic on?

Many thanks in advance.

B