When a machine that is configured for DHCP is initialized, it will broadcast a request for a lease from ANY DHCP server (DHCPDISCOVER). ALL DHCP servers that receive the request respond to the client with a lease offer (DHCPOFFER). The DHCP client accepts the first offer it receives and broadcasts a message specifying the IP address it has selected (DHCPREQUEST). The server that sent the DHCPOFFER responds with an acknowledgment (DHCPACK) and updates its database. When an address is renewed, the client will send the DHCPREQUEST directly to the server that gave the original lease. If the address is still available (which in your case it should be) the server will send a DHCPACK back to the client. Because this process starts at 50% of the lease expiration time, if the client does not receive a response it will continue to use the IP address it has and attempt to contact the initial DHCP server. If the lease expires or the DCHPREQUEST is met with a DHCPNACK (meaning the address is no longer available) the process starts from the beginning.
So. It does not matter what subnet the client WAS on if it is looking for a new lease. If both subnets are on the same layer 2 devices, any DHCP server can answer. Are your servers seperated in such a way that there may be latency or connectivity issues at renewal time? What I mean is that if all your clients on subnet 1 are on the remote end of the radio link and both DHCP servers are on subnet 2, any server is game. Routers in this case will not solve your problem because there is no routing to be done (nothing to route on layer 2) and any helper-address statement would just add latency. BTW, do you have the DOMAIN option set up in your DHCP server configs?