"As nice as it might be, I think stretching "preferred form" to "the form that would be most convenient for you" goes beyond what the license is trying to say."
It's not about what's most convenient for *you*, the downstream guy. It's about what was the preferred form for modification *by the upstream provider who modified it*.
This shift in perspective is critical.
The whole philosophy of the GNU GPL is to create a community of users for works of software wherein all are co-equal (within the limits of individual aptitude, of course).
That the one-big-monolith kernel SRPMs are a modifiable form of the work is not the point. As you implied, object files and preprocessor output are perfectly modifiable as well, but are seldom the preferred form for modification under the GNU GPL. These monolithic-patch SRPMS are not, by all accounts, the preferred form of the work for modification by those developing and distributing them--Red Hat's kernel engineers.
Congratulations are in order, I suppose, to Red Hat's management, who look poised to pull off what Larry McVoy could not--lock up a significant chunk of Linux kernel development behind a paywall without ominous notes of caution from the preeminent journalist in the field.