The positive point is that we now share same understanding.
The rather negative one is that this is far beyond my knowledge as with such design, I don't understand how devices on LAN could reach .2.1 and .2.250 interface on each network. Although this is not your current question, it makes me very confused about this global network layout(*) and potential side effects regarding your telnet related problem.
Just in case some changes occurred between 3.0 and 3.2, and as your hardware provides multiple NIC, as a matter of debugging process, I would suggest to split 10.37.2.42 and 129.37.2.42 on two different NIC, connecting both to same switch.
It will allow to set FW rules for each interface and have better tracking while running traceroute to understand what's wrong. Once solve, you may go back to your current design.
(*) to me, device laying between 2 physical network must have IPs on different subnet for each interface except if you decide, which is to me very very strange, to handle specific route to devices that are "on the other side" on same subnet.