However, RED like approach to attempt retries only a few times may be a good strategy to improve latency.
A RED approach would be good,
15 authors of RFC 2309 can't be wrong. :-)
but in general there has to be a limit on the queue. Your wireless interface should not become a packet long term storage facility. I've seen RTTs to the base stations on the order of 25000ms (that's 25 SECONDS) or more. If you can't deliver a packet a through 300ns distance of air in 10ms, it is probably time to drop it.
I think there is some middle ground between 25000 and 10 ms. While the former is definitely too long, the latter is probably too short. (Ignoring the fact that base stations may need to buffer packets for much longer than this when clients are in power save mode.) But the problem with sharing the airwaves is that you can't be sure how long it's going to take to deliver packets. The difference between first try @ 11 Mbps and having to retry several times @ 1 Mbps can easily be a factor 40. Last but not least, any system sitting between a high bandwidth link (100 Mbps ether) and a low bandwidth link (802.11b) needs to buffer to accommodate for the bursty nature of IP. Usually a round trip time worth of buffer memory is recommended. So that would be around 300 ms worth of 6 Mbps (= net throughput at 11 Mbps) = 220 kilobyte.