According to my test it seems MATLAB does not sum by chunk when operating on vector, to ensure the result is consistent, i.e. not depending on number of threads.
Elapsed time is 0.005212 seconds.
Elapsed time is 0.005153 seconds.
IMO opinion the sum just carried out linearly from left to right with some internal internat result with fiw number of bits > 64.
IIRC, in some version (2015?) MATLAB implements a multi-thread on vector and that raises some questions and that has been discussed on the old newsgroup, then they switched back to single thread.
The multi-thread is used for sum on 2D or ND-array, where each thread is in charge a set of vectors.
All that is hypothetic as TMW does not document the algorirthm.