On the one hand, that's not necessarily true. If you're working for a distribution but work closely with your upstream, you might work "in" your Debian source or SRPM environment, but test for the existence of a user-reported bug in the upstream code first, and if it is present there, patch it there before porting it back down to your own distro.
Needless to say, this pattern is often done in reverse, but I'm sure a lot of upstreams would like to see more instances of the above.
And that's exactly why Red Hat's move here has some kernel hackers (Greg K-H at least) wincing.
On the other hand, as long as there is no information loss either way in a two-way transformation, and the tools to convert from one form to the other are widely available, I don't think the spirit nor the letter of the GNU GPL is infringed.
That Red Hat's conversion from source+patches to monolithic when generating the SRPM is inescapably information-lossy is precisely the competitive advantage they are seeking from it.