What sort of tradeoffs would you make between true RAM latency and RAM capacity? At what capacities, CAS latencies, or speeds have you found diminishing returns in performance?