How much do you think it costs to buy a standard part, and hire a dozen engineers to get it all working??
My experience with embedded hardware type startups would lead me to think you would pretty much double your FCC registration process/expenses, add about 50% overhead to your QA budget to basically test everything twice, add about 20% to your hardware layout budget (unless the two chips are highly pin and form-factor compatible), and a few other incidentals. On top of the other extra basic engineering code to "get it all working".
Then of course you also have to WANT to have the unit on the carrier that you added the CDMA chipset for in the first place.
The harder part to gamble on in a situation like this is: how many people will switch carriers to get a truly hot device? If you get it right (as Apple has appeared to), a large majority of people will pick the phone over the carrier. I'm sure they would have sold more iPhones to date than the current number if they had a CDMA version, but I'm not convinced it would have been a truly significant number overall.
Anyway, that's my rough experience with hardware design. What's your perspective on the added costs?
Oh wait. They are? Huh. They must just have decided that the masses want iPhones. If they have to lose money on them... oh wait, you mean that isn't how business works?
The extra SKUs matter: if Apple is selling these phones directly (always a key part of their strategy, at least in the U.S.), how do they make sure that the customers buy the right one for their preferred carrier?
Training in carrier plans and pricing matters: How much training do they have to do with their employees about the different rate plans for different carriers? Configuring a CDMA phone on the cellular network is also more work than plugging in the correct SIM. The training differences involved are substantial and fundamentally against the way that Apple works: as simple as can possibly be done.
The quality of service/experience difference matters: would you care to explain to a customer why their buddy's iPhone on AT&T can browse the web while talking yet their "exact same iPhone" on Verizon can't?
There are lots of other extra costs that may or may not be covered by having a CDMA version that could reach a few million extra CDMA subscribers. Worldwide, there's ~142M CDMA EVDO subscribers as of 4Q 2009 (via cdg.org) and ~506M GSM 3G subscribers. The worldwide market for CDMA phones is ~500M, whereas the equivalent GSM one is ~4B. Apple made the right choice.
Here in Canada, Rogers's service doesn't suck (at least in the big cities), although their pricing sort of does (not enough competition). Bell and Telus switched to a shared GSM network (they're running both GSM/HSPA+ and CDMA until they can get enough of their customers switched to GSM phones to make it feasible to kill the older network). This was done mostly because they were preferred carrier for the Olympics this past year, and they didn't want all those Europeans roaming on the Rogers network, but it was done.
I personally predict that when Verizon has a major LTE rollout (which is essentially GSM), there will be a Verizon iPhone.