The CRAN could solve lot of problems of the cuurent architecture of communication networks to use the hardware in more optimized manner. This is defijitely more a valid case in urban geographic where the demand could be switching dunamically across different areas.
It's an interesting concept but I think the reality is that wireless networks are very competitive markets for equipment and for silicon. The standards and the silicon evolved together to optimally partition and solve the problem of wireless communications. It is unlikely that general purpose servers will be more cost effective than equipment developed specifically to solve the wireless network problem. Also, as you mention transporting raw antenna data over long distance fiber has problems as well. There are some dense urban geographies where consolidated Layer 1 processing may make sense, but it is highly unlikely that the wireless communications industry will be able to move into the "cloud" any time soon.
This seems like a natural progression and a nice way for carriers to cut costs. It's not too surprising that the chip venders aren't jumping behind it, because it really isn't a huge volume market for them. Moving to standardized motherboard architectures with PCIe cards for the application-specific cell network handling also could mean lower upgrade costs as new cell standards (i.e. LTE Advanced) come along.