Apart from your problem with the dev fee reconnect: You shouldn't use double hash with CNv7 on an AMD FX. Bulldozer has got 2 MB L2 cache per module. That's exactly the scratchpad size of one CNv7 thread. So don't us double hash, which results in 4 MB scratchpad size, use four single hash threads and put one per module, ergo (under Windows) cores 0, 2, 4 and 6 (or 1, 3, 5 and 7). That's way faster than your config. With CNv7 you should get a hashrate of approximalely 290 H/s.

The maximum you can get with an AMD FX is, to use the circumstance that it has – beneath its 2 MB L2 cache per module – an 8 MB L3 victim cache, which is not inclusive as it is the case with all Intel CPUs. Data in the L2 cache is never in the L3 cache and vice versa. So with bulldozer, you can start 8 single hash CNv7 threads dedicated to the cores 0, 1, 2, 3, 4, 5, 6 and 7. This config will result in a lot of traffic between L2 and L3, that's why performance will not scale linear with the number of threads, but still all data is cached, no slow memory access. It will give you the max. performance, in your case around 340 H/s with FX-8320.