We've gone slightly off point, but to answer you questions specifically on software:
Yes, you do see Google sorting out x86 compatibility (and MIPS), on several levels. The explicit support for x86 in the VM runtime (java) and the support for x86 in native applications in the Android toolchain (compiled by the developer, ie, right click, target x86), support in the play store, and finally Intel has provided the emulation later for a fall back when the developer hasn't produced a native application.
Contrast this with MS, they provide a full Windows 8 environment with RT which is locked down so it can only run their windows applications (office). While you can jail break RT, to run other windows applications, legally the ARMy can't 'fix' this, even if it wanted to.
I'm not sure where you get the idea that there is a porting software issue for ARM on servers (Linux)? While I work in a 100% x86 shop, I can tell you categorically there isn't a problem running the workloads that the ARMy are touting for. Yes, you wont be able to run MS SQL server, but the LAMP stack, no problem. Even some of the edge cases were there is a dependence on x86 assembly (ie, compilers or SIMD intrinsics) I've seen explicit support for the ARM ISA.
Suppose you're buying equipment for your server farm. You could buy x86 based servers and simply know that it will run everything in the Linux/Unix/MS world, or you could buy ARM servers to use on the subset of Linux applications. Unless there is some huge cost advantage upfront (which is not likely the case) or operating cost advantage (also not likely the case) it would be unwise to buy the equipment that is limited in its use. And later, when one is ready to upgrade to newer gear the ARM-based equipment will not have the same value in the secondary market because of its limitations.
Total cost of ownership includes the residual value of the gear when traded and the ARM gear is likely to be less favorable to the owner.
It's a bit like Windows RT tablets. Yes, it can be built, and yes it can run some applications but there's no compelling reason to buy it given the availability of more versatile products at comparable cost.
I think you misunderstand the target that the ARMy are going after (BTW, just because I am discussing this, it doesn't mean I think the ARMy will win any meaningful share). When you buy gear for a new deployment, you always test it. You don't buy 10 million worth of severs and suddenly go 'oops', xyz doesn't work.
Perhaps, it's the organisations I've worked in, but the second hand value of the gear was never factored into a ToC cost as it was always scrapped:)
If your workload is CPU bound or need low latency, you wont go ARM.
If you are a typical corporate data centre, you wont go ARM.
If you workload is generalized (ie, a mix of workloads), you wont go ARM.
If you are typically IO bound, and running a specific large scale workload, then ARM could be for you. We're talking things like Amazon AWS, S3, Cloudfront and ElastiCache.