Lightning Edit: Goons (tuned for the matchup) might be the best BM against this?()
I feel Cultist BM should also do well (admittedly, this is an unqualified guess - I have never played with Gear).
Well, yes, I was going along with the "non-junking" mentioned in the OP.
Is there any intelligence for what to set aside?
Yes. As assemble_me notes, you can go a long way by simply saving your actions and money that isn't helping you get to a price point - and I believe this is how Geronimoo programmed his simulator (meaning it's enough to get these stellar results). At the same time, I think this can be improved upon a bit. You point out a very obvious example (though I think the clever Mr. 'moo got around that one a bit by requiring enough money in deck before greening, which mostly corrects this problem). Still, of course, an endgame sense of what VP cards are important right now is going to help a lot in the endgame (saving more to obey PPR makes the rule a bigger deal here than in normal BM, but also sometimes if you have $6 or something, the Duchy has little equity over an estate, etc.).
In general, there are a few things you want to be conscious of: first of all, you want to know as best as possible what your next hand will hold. This will help you sculpt out your next few turns now. Also, keep track of when shuffles are happening. You get way less equity from saving good cards if it means they'll miss the shuffle, and you get some positive equity from saving bad cards (green in particular) to make them miss a shuffle. You can also sometimes save nothing so that your Gear itself doesn't miss the shuffle.
Furthermore, these things probably get more important as you start adding more kingdom cards that might be useful. Ironmonger is good and better than silver, but enough to not save an extra money? And Venture is good but worse than Gold - how much worse, really though?
For just a little taste of this, let's look at the case where we open Gear/Gear and draw a Gear on turn 3 (as will happen 68% of the time). If you draw to 6 coppers, you're going to buy Gold. Ok, that one's easy. But it's worth noting that in every case, you know you will have your other Gear next turn, that it will trigger a shuffle, and indeed you will know the exact cards you will have in that hand before playing your Gear. There are a lot of combinations of 6 cards you can have, but after giving a lot of thought, I actually think what you want to do, if as-it-usually-will-be, it's possible (with the only exception being 6 coppers->gold), is to make your current hand 1 estate and 3 coppers, buy a Silver. This makes your next hand 4 coppers, 2 estates, and a Gear. You'll play the Gear, which will trigger a reshuffle, where the only cards are Copper, Copper, Silver, Estate. You have a 1/3 chance of drawing Copper/Estate, in which case you save 2 coppers and buy a Gear, a 1/6 Chance of drawing Copper/Copper, in which case you save nothing (I think) and buy Gold, a 1/6 chance of drawing Estate/Silver, in which case I think you save 2 estates (not 100% on that, might be better to save nothing) and buy a Gold, and a 1/3 chance to draw Silver/Copper, in which case you probably save Copper/Estate and buy a Gold (but this is the hardest for me to work out). Following that, you are buying Golds about every turn, though you want a 3rd Gear if you don't have one and brick off (which is more likely to happen if you only have 2 Gears, so it works out nicely). And then you can green.
It's actually a bit startling how far you can plan ahead from so early a point. Really, you're taking out most of the luck except for whether you get a gear in the top of the shuffle or not - in some ways, it's kind of like an engine. And the 'secret' is that if Big Money draws perfectly, it's actually very strong. Big shocker
(edit: fixed an obvious small math error - counting is hard)