Then this question boils down to: would huge latency (or if the software is aware and allocates properly) added by having cpus on 2 sockets, but having 8 actual physical cores beat 4cores+4hyperthreadcores on faster mem.
If the software is optimized for dual cpu/8cores, in fact I do think that 8 physical cores would beat 4real/4hyper cores, specially considering i7 will only be 1-2% faster going from 1600 to 2133 mhz for example.
In a perfectly optimized software, (and seeing encoding is one of those cases where multithreading actually does work well) 8 real cores having more total cache would definitely beat 4/4