VMware's policy is not to comment on future features/products so you will probably not get any direct answers here.
You should reach out to your local VMware team to get a roadmap update from one of the SE:s.
Its correct that Citrix can do this with the Nvidia VGX technology (its more Nvidia then Citrix to be honest...), but then only XenServer is supported and you can then carve up the GPU:s in smaller chunks and use Nvidia driver with DX11 support.
My doubts about that VGX is that the GPU:s will maybe not be able to satisfy the users since they have to share them and you can over-allocate the resources.
(A K2 is basically 2 K5000 on a single PCI:e card so depending what your users have today it might not work for them to share one K5000.)
Would a few K1:s be a possible solution for you? Then you could double the amount of users per box compared to K2.
Another option is to use lower performance GPU:s in vDGA like the Q1000M/Q3000M or the older fermi-cards like Q4000.
// Linjo