Since Marvell had the highest throughput, I thought perhaps they'd also have the neighboring WLAN thing licked. I found that they don't, but there is something different going on in Marvell's rate adaptation algorithms.
Figure 11: Neighboring WLAN Test - Netgear RangeMax Next - Gigabit (Marvell TopDog) - uplink
Uplink performance shown in Figure 11 looks like more of the same. But Figure 12's downlink plot is the first evidence I've seen of draft 11n products actually downshifting out of channel-bonded speeds.
Figure 12: Neighboring WLAN Test - Netgear RangeMax Next - Gigabit (Marvell TopDog) - downlink
Figure 12 shows that the Netgear RM Next - Gigabit WLAN starts out at very low throughput at the tail end of the first 11g WLAN activity and continues at a very low rate even with the 11g WLAN off the air. But when the 11g WLAN starts up again, something kicks in and the Netgear WLAN finally shifts into 40 MHz mode and crushes the 11g WLAN.
Figure 13: WN511T Client Utility - 11g WLAN active, draft 11n idle.
Figure 13 confirms that the Netgear client is reporting transmit and receive link rates of 2 Mbps (which is actually an 11b rate!) during the 11g WLAN's initial transmission. I also found that for the run shown in Figure 12, the link rates remained at 2 Mbps when the 11g WLAN first went off the air. But when I repeated the test, I didn't find consistent behavior.
Figure 14: TopDog being a little too neighborly
Figure 14 shows another variation, this time with the Netgear pair using a 6.5 Mbps transmit rate (which is actually the lowest 802.11n 20 MHz link rate) during the low throughput period. Unfortunately, this behavior penalizes the Marvell TopDog fueled WLAN too heavily, which I'm sure is not the designers' intent.