Aren't SYN-cookies [0] the traditional defence against SYN floods? The proxy, or a device in front of it, could take care of that. Only connections that complete the 3 way handshake would take up any room in the connection table.
I don't understand. Why is it simple to filter UDP but not simple to filter based on a cookie? Is the validation cpu-expensive in bulk? Do you not have the capability to filter that way at the boundary?
0 - http://tools.ietf.org/html/rfc4987