Yes this is the conclusion I have come too. I believe Energia' implementation of the Watchdog here is what causes the higher usage. It is very impressive that it manages to get even close. I think it is abit misleading to call SleepSeconds() true LPM3 though.
It would be great to see some other implementation though; as the MSP430 is capable of sub 1ua sleep.
Perhaps I will experiment with a library in the future to make this possible.
Thanks for your response.