Ted T'so showed up here on HN in one of these discussions and derided any criticism as "ivory tower" and "academic".
OTOH, Kroah-Hartmann and Torvalds seemed to approve of my essay ("Myths about /dev/urandom"), so maybe not all hope is lost.
Still, let's not lose perspective here: the Linux random subsystem certainly has its flaws, but it is secure for all practical purposes. Just not convenient.
It seems you would want to slowly change/update the entropy pool to protect against compromise. So can there be a worst-case situation (if you don't block - for the different reason) that your entropy pool is not updating fast enough to recover from compromise. Or the pool update requires so little entropy (compared to what linux is doing with kind of refilling it) that its almost impossible for that to matter?