When I run msl12 on a modis image I notice that over large parts of my scene I have nonzero La (my test for an ocean pixel) and zero TLg. Is there a TLg threshold below which it's neglected? I'd be surprised even if so, since as I understand it MODIS doesn't have the tilt that SeaWiFS has, so I'd expect greater sunglint.

The glint radiance is determined via the Cox and Munk model.  No glint correction is performed (TLg=0.0) when Lg/F0 < 0.0001, where Lg/F0 is model radiance normalized by irradiance (ithis is the msl12 product known as glint_coef).  The l2_flag MODGLINT indicates glint > 0.0001.

This is not to say there is no glint, but the model predicts no significant glint for the geometry in question.