The latest git version of rt73usb gets into a tight loop which according to top attempts to take up all available CPU time. It appears to happen almost as soon as the stick is inserted, that is, before any attempt to associate takes place.
Aside from the fact that it takes around 95% CPU and slows the machine down, it appears to behave normally. That is, it will associate and communicate with the outside world, and nothing unusual is reported in the logs.
As soon as the wireless interface is brought down, CPU usage returns to normal - it is not necessary to unload the rt73usb module.
I get the same with the latest wireless-2.6.26 git, but not from 31st March, so judging just from the timing of it, it appears to be one of four patches
* Remove MAC80211_LEDS dependency
* TO_DS filter depends on intf_ap_count
* Invert scheduled packet_filter check
* fixup some non-functional merge errors
Could you try to revert any of those patches to see which patch introduces the problem?
You can exclude the patches
*fixup some non-functional merge errors
*TO_DS filter depends on intf_ap_count
since those patches are quite trivial and couldn't cause the bug.
The other 2 patches could both be a probable cause..
It's "Invert scheduled packet_filter check "
I can confirm that this deals with the problem.