Well I just found this thread... I have been running a machine with 1 9800gtx+ and 1 8800 gt for a couple of months now and as of last week I can only get 1 wu at a time, for the 9800. The 8800 no longer runs. Bummer. Since I can't run it anymore I will retire it. I just can't afford to buy $1800 of cards at a time. My car is worth less than that......

Had this been planned for you could still get work for the 9800 from here and for the 8800 from Collatz and the system would assign the tasks appropriately ... and that is something you still may want to try ...

Searching through MW forums, about new ATI / AMD GPU 5870\5850, was found quite interesting message posted by Verstapp.
Lets look, what he was found & shared with us:
http://www.theinquirer.net/inquirer/news/1557395/nvidia-bastille-stormed-hackers
Sure, this could add some thoughts.

Which is not the first time that ATI or Nvidia did something like that ... yet, Rom says that the UCB expectation is that mixed ATI and Nvidia installations are going to be easier to support ...

An idea I think UCB should consider: Support a list of GPUs, each with a database section describing the capabilities of that list item. Allow including clusters of GPUs that meet their expectations for matched types within that cluster. Include a mark on each one saying whether it is free for use, already in use, or possibly the user has marked it as not to be used now.

Give the CPU parts of workunits the capability of inspecting the entire list, and choosing at least one item on the list but not currently in use as something it wants to use, returning something saying there is something on the list it can use but it wants to wait until at least one more item on the list becomes free to use, or returning something saying there is nothing on the list it can use. If the CPU part of the workunit gets access to more than one item on the list, it then has the responsibility of dividing up the GPU part of the work among the different items, and possibly freeing some of them for other uses early if it decides not to use certain items this time at all.

That way, the CPU part of the workunit, and not BOINC, gets the responsibility of deciding what GPUs it can use, and can decide whether it is capable of using unmatched types at once.

Also allow the CPU part of the workunit to decide that that it wants to take responsibility for deciding how long to wait for something currently marked as not to be used becomes marked as available for use again.