It is quite sensible to cache results, indeed. However, you do run into a problem of the constant changing nodes. For example, after a few minutes, your PUSH route (for firewalled nodes) may already be lost. After an hour or so, the node that had the sent the result may have gone offline too.