If you are connecting the blade directly to the pool with the getwork protocol, try instead to set up a stratum proxy and connect through that. In that case the hashrate will be exactly the same.

The stratum proxy will be generating new work locally and the blade hashing it, so the blade should be running full speed all the time. If the hashrate is different then you probably didn't measure it long enough to even out inaccuracies.