Except now you're loading the emitter with 2k to ground. Your 100k resistors in the circuit become pointless since they're in parallel with 2k, and the 2k and 3.9k on the collector side of the transistor becomes a voltage divider that drags the transistor down towards ground, affecting the bias.
You may need to add a pair of buffer amps to keep from loading down the circuit, turning the difference amp into an instrumentation amp. With a gain of 100, you're going to have to watch out for op-amp offset voltage and drift as well. The TL072 has a 3mV offset spec, multiply that by the gain and you have +/- 300mV of uncertainty in your result.