Several reasons I could think of off hand might include:
ARM is an especially active and interesting architecture due to widespread use. Meaning that a lot of effort is going into compilers for it and porting of more common applications will be along for the ride. Given that one of the focuses of the product is for desktop use under Linux, this is a significant factor.
Also since ARM is popular it has more resources for basic development. We see this in the continuing improvements in die size and related efficiency for instance.
They seem to be expecting to employ many cores per chip. I've long felt that this will be the most likely way forward for a whole range of hardware classes and have been surprised that it's taken as long as it has to come into the fold. It complicates many things and is sub-optimal for certain kinds of problems, but eventually it's probably going to end up eclipsing the limited core architectures on a lot of fronts. Biting the bullet now will very possibly give one a leg-up on their competition in the future. Of course SPARC, at least, is conducive to doing this as the higher end SUN gear has demonstrated for a long time, but I doubt that that architecture will ever be as competitive as ARM in the lower end power sensitive market segment.
Lastly, if they do actually intend to open-source some of this stuff it would be a complication if they start with a processor which is in widespread use in the military industry.