Comment 249 for bug 134660

Found some relevant comments about why the speed of the connection may be lower than expected.
http://lwn.net/Articles/278191/

To quote it here:

"At the moment the semi-free madwifi is more reliable than ath5k with 2.6.25-rc kernels[*], due
(I suspect) to problems with the mac80211 layer being rather too aggressive in reducing the
link speed in face of less than ideal signal conditions. Signal quality/level monitoring
(using wavemon) suggests that unsmoothed values are causing the rate algorithms to believe
that conditions are worse than they actually are. I see similar problems with the in-kernel
rt2500 driver."

And in another comment from there

"It appears to be tweakable via debugfs." - http://linuxwireless.org/en/developers/Documentation/mac80211/RateControl/PID

Hope this is helpful to some ...