It's good to hear that you can step up to ATMega8 processor for less than 5% cost increase, because 8k chip can be programmed to do so much more and you can add more value to your product.
If you would like to share more details about your project on this forum you might get a few suggestions and some you could find useful.
For instance, what is the main chip you are using for the board?
What are the other parts that contribute significantly to the cost (motor drivers, regulators, connectors, crystals, etc...)?
What are the PCB costs, i.e. setup fee, area fee, per hole drill costs, board cut costs (i.e. rectangular vs odd shapes)?
What are your general battery power objections?
How close do you want to stay compatible with Arduino (software only or pin layout as well)?
I bet many of us have seen some crazy to ingenious ways of cutting production cost and increasing product value at the same time.
Figuring out lowest production cost is a great challenge and fun for any engineer, and it sure is to me
1. If PCB area cost is cheap and it is cut for you, just lay copper traces for USB connection on the edge of the board with proper cutouts and you don't have to solder USB connector.
2. If you choose ATMega8 chip to be your USB-to-TTL bridge and you are going to sell it in every robotic kit, then put it on the same board and avoid having second board and any connecting wires. Just run RX/TX/Reset lines via jumper connector or wires, so that if necessary, end users could solder in header and connect either side to something else.
3. Once both chips are on the same board, they could share the crystal and it will free XTAL2 pin to be an extra I/O line on the main chip.
4. In general once you commited yourself to having two microcontrollers on the board it opens up whole lot of possibilities. You just have to make PCB design open enough that they could be realized without too much hassle.
5. ATmega8 chip could be have code to be ISP programmer as well, just make traces and holes for optional connectors.
6. ATmega8 chip can do TV video out. Again if PCB area cost is cheap you could leave a small prototyping area for several resistors and diodes and it could be figured out later if this thing is doable or not.
7. Presumably, at the later time special code could be developed for the ATmega8 USB-to-TTL chip that will on reset talk to Arduino IDE pretending it is a main chip's bootloader and then program the second chip over SPI. This frees up extra 2k of memory on the second chip. Is it usefull? I don't know, but posibility of being able to do something like that adds value to the platform.
8. When USB-to-TTL is not used ATMega8 could serve as I/O extender to the main chip either via UART, I2C, or SPI bus. And with proper software mods even while doing USB-to-TTL. This adds value to the entire platform.
9. What is your mechanical platform? Mine is Lego. Doing something as simple as drilling m3 holes that are both on 10mm and 8mm grid (for example 40mm apart) make it instantly Lego friendly. If your motor drivers could pull 0.6A at 9V you electrically compatible as well.