Changing only master could make backports difficult to impractical.  (Pure module name changes would be an exception if gits keep track of them and reverses them for a backport.)

If a module is touched, user changes to that module get wiped out when the module is replaced.  Aside from that, most of idlelib was declared private and 'subject to change' to allow refactoring.  The only problem I have heard about is that 3.6 broke IDLEX (as I expected).  IDLEX which made changes far beyond the narrow extension protocol.  Roger solved that by incorporating a private copy of 3.5 IDLE with whatever modifications he wants.  This is exactly what Nick said people should do, in the 3.6 alpha discussion, if they wanted frozen IDLE.