Puzzling XFR question

[slidge at slidge.com]

 I have a nameserver system that has a hidden primary (ns0) and three
 slaves (ns1, ns2 and ns3).

 I'm having an issue with ns1 and I would like to see if anyone here has
 seen this behavior before and has some recommendations.

 ns1 is having difficulty transferring zones from ns0.  Currently I have
 21 zones.  If I update the serial #'s on all the zones on ns0 and issue
 an "rndc reload", ns2 and ns3 transfer over all the zones almost
 immediately, but ns1 takes longer (at this point, it has been 37 minutes
 since I reloaded ns0, and ns1 has transferred over 15 of the 21 zones,
 including five that I did manually via 'rndc reload <zone>'.

 I checked network connectivity (ns0 is in upstate NY, ns1 in Virginia)
 and there seems to be no issues.  There are also no firewalls in between
 the nameservers.

 Restarting the nameserver on ns1 did not cause ns1 to go out and
 transfer the zones any quicker, nor did issuing an 'rndc reload' on ns1.

 However, issuing an 'rndc reload <zone>' causes an immediate zone
 transfer for the specific zone in question.  I can also do a manual zone
 transfer using 'dig' with no problem.

 I am running BIND 9.2.3 on Solaris 2.8.

 Anybody know where I should start looking to troubleshoot this?

-- 
slidge