It's a relatively well-known thing that wifi speeds decrease if you're too close to the access point: the "screaming in your ear" effect. But we now have stats that demonstrate it clearly.
I zoomed into just Chromecasts. In the top graph, notice the speed decline starting at -30 dB or so RSSI. In the second graph, notice how there's actually a spike in number of devices starting at around -30 dB.
My guess is the bimodal distribution happens because there are two places to put your chromecast: right next to your AP which is right next to your TV, or not.
We talk to the chromecast on 2.4 GHz, which means we're using ath9k. The ath9k driver currently sends all packets at maximum transmit power, even to devices right next door. I wonder if a minstrel-like algorithm (such as minstrel-blues) which experiments with different power levels would actually fix this problem and achieve full speeds (downstream toward the chromecast anyway, albeit not back upstream) even when gratuitously close by.