Stable without fans forced on max running AMD NBench 3, temperatures stay reasonable. I'll probably leave it here.
Rightmark CPU Clock Utility log while benchmarking:
00:05:01.984, CPU: 0, Clock: 2687.47, Throttle: 100.00, FID: 12.5, VID: 1.350, CPULoad: 89.47, OSLoad: 88.89, CoreTemp: 55.0
00:05:01.984, CPU: 1, Clock: 2687.38, Throttle: 100.00, FID: 12.5, VID: 1.350, CPULoad: 48.10, OSLoad: 46.03, CoreTemp: 58.0
Lowered Vcore to 1.35V at 12.5x and 1.2V at 6x. (1.1V is too low at 6x, caused hang)
Small but measurable improvement with Blender, rendering a model of a Dodge Viper RT-10:
(pod = performance on demand profile, others are maxed)
FSB*MUL = MHZ = Mem Ausmarks, Cpu Ausmarks => Blender render 1 frame
215*12 pod = 2430Mhz = 630,8, 1188.1 => 3.67
215*12.5 = 2687Mhz = 635, 1230, 3.48 => 3.36 5.5% improvement
215*13 = 2795Mhz = 661, 1264.5 (13 = heat issue, 12.5 = none such)
215*13.5 = 2902.5Mhz = 664,1307 => 3.40 (temperatures graze 70)
215*14 = bluescreen

220*12 pod = 2640Mhz = 647.2,1221.6 => 3.63
220*12.5 pod = 2750 = 650.8,1250.5 => 3.53
220*13 pod = 2860 = 678.5,1301.3 => 3.4 (temps rose quickly)
nBench3:
Baseline 200*13 2600Mhz: Integer-1 3348 Integer-2 3263
Overclocked 215*12.5 2687Mhz: Integer-1 3430 Integer-2 3263 (no change)
Summary, I'd just say that ramping up the FSB does demonstrate minor performance improvements even if the CPU multiplier must be dropped to compensate.