A bit off-topic but what's a reasonable getwork interval anyways? Shouldn't the client complete the work before asking for more?
if he would, he had to wait for the result of the getwork - time for your gpu to cool down
as (probably) you don't want that its better to prefetch some.