Frankly I don't understand why this exactly makes a difference,
but at least I can confirm from testing that the b44 part really
does fix the issue. Without the patch any network connection
is reset quite often (even loosing pings) and the PHY is powered
down during reset. With the patch applied network connection is
stable. I think the b43 I have in that machine never worked too
well with the b43 driver (had to use wl), so I cannot be sure
about the b43 part, but at least it is doing the same changes as for b44.
Eric sent out the patch to be added to linux-net, but I am not sure how
long it takes to get into mainline. And since it is a regression and
I was able to test (most of) it, I think it should be added as soon as
possible.