I wonder if we can achieve a better accuracy when aligning with Meshlab. Although it's more time consuming.

Exactly Hugues, my thoughts entirely, and it's also important to move the object to get most information in the first pass, which is harder to do on a single axis turntable.
What Nerv and others have done with Free Scan mode backs this up, and you really want to progress your scans to provide non ambiguous information to the cameras, that way the self align function doesn't get confused.