While the unknown g might or might not be amenable to this approach, it's really fast on g == 1:
Expectation[
1, {x1, x2, x3, x4} \[Distributed]
MultinormalDistribution[muvec, sigmat]] // RepeatedTiming
(23000 times faster than NIntegrate.)
Then there's also NExpectation to try, too.