I'd guess the problems were more software than hardware. The only hardware interface required is for cipher initialization with the SIM card, which is identical to the process the hardware must do for standard cell towers. The UMA specification is surprisingly complicated, and there aren't too many software stacks for it. If it's going into all of T-Mobile's future Android handsets regardless of actual manufacturer, the UMA implementation could well be part of Android. Thus, the implementation needs to be acceptably licensed so that the multiple hardware manufacturers can use it. I doubt the existing companies with UMA software stacks would be willing to license something like that in acceptable terms.
I bet the implementation is probably brand new from the UMA specs, and it's just taken this long to complete and validate. If that's the case, a new radio baseband and software update should be sufficient to bring it to existing devices.
