Because it is too smart for its own good
The SIP itself will work though PAT in most cases it is the actual voice going between the 2 device that does not. Although it is sorta a over simplification the 2 end device form a connection via the SIP/H323 control. This is a management session used to negotiate things like codecs and other things. The most important are what the 2 end ports are to be used. Both ends sorta pick random UDP ports to run the actual voice session on. They tell each other the ports to use inside this management control session.
The end devices then just blindly send voice data on these ports since it is UDP based in most cases. Since the PAT device has no knowledge of this "secret"...