This gets asked often enough that I wrote down my notes.

1)The laser relative intensity noise sets a floor that no amount of RF gain can fix.
2)Shot noise at the photodiode scales with photocurrent, so more optical power helps up to a point.
3)Thermal noise in the post-detection amplifier dominates at low photocurrent.
4)For a given link, there is an optimal optical power that minimises the total noise figure.

YMMV depending on the project.