Thanks for all the input!
I ultimately decided on g6, which the computer analysis likes (it considers it the best move and rates the position as -0.01 after it), but I really hate in retrospect. It's one of those "passive, but that's okay because we can play perfect defense for the next 10 moves" sort of computer moves. Stuff like this is why I've almost completely given up computer analysis for my games.
Two moves later we've got:
And here I think black's queenside play could materialize. Nc5 and black starts to get some nice room to maneuver, while it's going to take white forever to get anything real going on the kingside (example: Nc5/Bc2 Nxa4 and white's got to concede the bishop pair and black gets some files to play with).