Thanks for all the comments.
Attached are a couple of updated drafts.
I tried to lighten the map up overall. By using smaller gradients, and shifting them to the lighter side, I was able to remove some of the darker colors, and, I think, the need for the halos on the labels, which makes me happy. I also ended up lightening the corner quite a lot on the overview map – that's actually the continental shelf, so depth drops dramatically (might as well be a cliff at this scale), so I opted for essentially saying "everything over 250 meters gets the same color" even though the bottom of the shelf is at least 500m down. It makes it noticeable that the shelf is there while not making it so overwhelming.
Instead of using the 50m isobath, I instead shifted to making the transition between 50 and 75 meters have a higher contrast than the others, which I think emphasizes that transition, while not being as overpowering as the white line.
In adjusting the graticule, I also discovered that while ArcMap will happily let you assign a stroke weight below .25pt, it doesn't actually export it as such to the PDF, and in fact if you assign it say, .2, it will end up displaying as 1pt. Which is why my the graticules were coming out much heavier than I expected (and making them a too heavy, as others pointed out).
Addressing specific comments:
David: the title doesn't say "Snow Crab Adult Females" because the whole article is about snow crabs, and there will actually be a series of maps showing different variables, so saying "Snow Crab" on each one seemed redundant. I took your suggestion on the titles. I don't know how I managed to overlook that.
The_Skua: I agree about the legend being a bit clunky. It's my least favorite part of the whole map. The reason for doing classed quantiles was to make comparisons across the series easier, following Cindy Brewer's advice in "Evaluation of Methods for Classifying Epidemiological Data on Choropleth Maps in Series" – DOI: 10.1111/1467-8306.00310. However, I appreciate that that article evaluates choropleth maps, not graduated symbols.
Essentially, I'm relying on the quantile classification scheme to do the normalizing of variables for me. I willing to be dissuaded, though. :) And you're right, five classes probably makes more sense than six.
I like your idea of trying to do something a bit more clever with the labels, though. I'll have to see if I can coax ArcMap into doing something nice.
I was actually unsure about the "Chinoecetes opilio commercial fishery" myself, and had to ask for more clarification. I think the new version shows that much more effectively.