I feel like t3andy is... overvaluing his own views on this. I am currently in the process of bringing a T3.1 based device (custom PCB using PJRC Mini54's) to market, and while it is a lot larger than what is being proposed here (my PCB is 280mm x 220mm), a lot of this stuff applies no matter what.
For one, everything Potatotron has said is spot on. Having no concept of what your user base might be - and I'm pretty sure a poll on this site will bias your numbers quite a bit - a SMALL run is your best bet at first. Here's the rub with that, though. Let's say, best (worst?) case scenario, your device takes off! Suddenly you have thousands of orders coming in. But you only made 100 initial units, and you didn't even give yourself enough of a margin that you could use the profits from those to build another run of 100. And you've got like, 1256 orders that are just trying to give you money. Well, now you have to rework your workflow a good deal - hopefully things are scaleable when you set it up the first time.
But oh shit, there's a problem when you went to manufacture your next run of 1400 units. The assembly house put the wrong resistor/crystal/capacitor/whatever (I had an assembly house put a 'stock' digikey MINI54 on a small run of my boards... partially my own fault, but I thought I'd documented it well that I NEEDED the PJRC ones).
So now what, you've got 1400 bad PCBs. You could spend a bunch more money, cutting into your margins, and send them aaaallll back to the assembly house to be fixed (maybe you'll be lucky and they'll give you a deal on the rework, especially if it was your own fault the mistake happened). Or you can try and do them yourself, which takes time - and you've got 1200 customers waiting and pestering you for the product they ordered, at this point, months ago. Or you hire people (money money money) to help with the rework.
My point is, even just sticking to 100 units, and hell let's be bold and say that you're REAL good with a hot air gun, so you're doing them by hand. What if you ballsed up an order of boards? What if the fab didn't understand your notes, and didn't castellate/copper the edge pins? We're talking a $33 device in the *best* case scenario. Thankfully, I think you'd be comfortably outside the purview of the FCC (kit parts aren't covered), which is what I'm dealing with right now (EMI shielding and design is both wonderfully fun and also a nightmare).
My real point is, if you really need a 'board' that you can surface mount, assumedly because you have other components you'll be surface mounting... Why not just build the T3.1 section direct on to your board? I've done a few by hand. It's not easy but it is doable. Hell, it's not that much extra trouble (all you need is a heat gun) to drop a SMD header pin section on the T3.1 itself, for the SMD pins, and design your PCB with those extra socket holes in mind. It's all of what, 2 extra minutes per unit and maybe $100 for the heat gun. Definitely not worth the extra $12 (at minimum) per unit to get this custom 'stamp' board made for a small run. And for anything over 50-100 units it's not that terrible a cost to just have the SMD components put directly on your boards at an assembly house.