That's a fun math exercise Mark. For the worse case 1/8 inch water thickness or 1/4 inch of path length at a sodium wavelength of .441 in water gives 14,399 waves path length (my earlier math was off I think). If we want no more than a 1/10 wave error then cos(14399/14399.1)= .999993 or 0.21 degrees half angle. That's about 45 feet or 134:1. Of course 1/8 inch is a bit much but this distance is proportional to the water thickness and mirror size. I guess you could set up a telescope at the right distance away with an ordinary mirror folding the path down to eliminate the collimating lens. Might work.