Chris:
Contrary to popular understanding -- and most Cisco documentation -- VTP client switches do originate their own VTP advertisements and propagate them out trunk interfaces every 3 minutes (using a 3550 running c3550-ipservicesk9-mz.122-35.SE5.bin) during stable times, and also every time a change is made to the vlan database.
If the client switch receives a VTP update from the VTP server, it will repackage the advertisement and place its own system MAC address in the source field of the MAC header.
If the trunk that connects the VTP server to the client is disconnected, the VTP client will no longer receive the adverts from the server switch, but it will still originate its own VTP advertisement and propagate it out its trunk interfaces. The source MAC of that update will also be its system MAC address.
So, if you are determining that the client is "switching over" to server mode when you disconnect the VTP server by the fact the the client is originating its own advertisements, this may be the explanation for it. The switch may not not necessarily be changing over from client to server mode, per se, but simply "acting" like a server by originating its own VTP adverts.
This is why I have always said that VTP is nothing but a colossal collection of confusing kaka -- and if at all possible, should never be used. It causes more problems than it solves and creates the kind of confusion that can knock out a good part of your network.
My 2 cents....
Victor
PS It is in fact a Cisco best practice to configure more than 1 VTP server on your LAN, but watch out for bouncing trunks....
