Have you considered getting an STM32 board? They're relatively cheap for what you get in terms of ram, program flash, and I/O (although all you'll be doing is reading values from an ADC or I2C and sending it 'home'):
They're approximately $36 a pop and have a built-in Ethernet controller and excellent tool-chain support (GCC).
Alternatively you could try and get a couple of wireless sensor nodes. I like the mc13224v-based redbee motes:
They are $43, but you get a nice, small form factor, wireless solution. Open source code examples available here:
The advantage of wireless is that they are non-intrusive, and you can set up a multi-hop network easily (i.e. using contiki: http://www.sics.se/contiki/) and you can power them with two AA batteries.
Finally, if you're going to run wires anyway, why use Ethernet at all sampling points? You can hook up multiple I2C temperature sensors to a single microcontroller, and just run the I2C wires instead of UTP. Alternatively you can have a single MCU at each sampling point (for example a http://www.pjrc.com/teensy/ or some other inexpensive little board) and connect them with low-speed I2C. Then, have only one of them interface with your server over TCP if you wish. Not sure if I2C is good over long lengths of wire, but there must be other interconnects that you could use. The idea is that you can simply build a minimum spanning tree and do simple store-and-forward networking to get data packets to the sink node (the one connected to the wired network).