Yeah that bottleneck will not even matter compared to those 8 threads of raw power while storing those instructions in that 8 MB cache
The 10Mb of L3 on the 3820 and 4820k is often overlooked as well. Consider the bump over the 2600k that the 3820 had, and it wasn't clock speed that boosted it. At the time, it was faster and even today it keeps up with most tasks you throw at it. Additionally, an enthusiast CPU doesn't mean it's guzzling down power if it's idling. My 3820 does pretty well on power all things considered, where the entire rig idling (with drives spinning) is about 175-watts. Add 100-watts under full CPU load, then add another 100w for each GPU under full load which is unrealistic in its entirety. So considering how much time my rig sits idling and sleeping, it's really not that much. My cloth dryer on 208v split phase probably consumes twice the amount of power than my tower does in a month's time. My tower alone is probably ~15-20$ a month of my electric bill and that might even be on the high end.