You are right, I should have mentioned that: 80-200 atoms.
By the way, we plan to run Gaussian and CP2K calculations on this machine as well.
I should also mention that we already have a workstation with the setup described in 1 (Xeon 5690). However, I am not really satisfied with the performance. While the single core performance is great, the scaling with VASP is OK only up to 8 CPUs (i.e. 4 "real cores").