my questions would be mainly targeted to AMD insiders, but if somebody is more well informed than me, they are more than welcome to answer too.
1: Will linux catalyst be detached from xserver?
When I first heard that the driver is somewhat merged with xserver, it already seemed to be an annoying "feature", but now it seems even more like it. NV driver can reach device resources without anything else. My aim would be to integrate GPUs into a LARGE grid system, but the problem is that all machines (especially the worker nodes) are running SLC5 (ain't too hard to guess which grid I'm talking about). SLC is a simplified Red Hat distro, but they try to avoid installing the graphical interface wherever possible, and WNs are the ideal exapmle of machines, that typically have no GUI installed. I know there are workarounds for exporting DISPLAY envvar to reach all GPUs, to avoid the DISPLAY variable interfering with bringing x11 windows across ssh, but these are all workarounds. There is no real need for the driver to depend on xserver. Please, try to solve this problem!
2: Threadsafe modification of fglrx?
It is still not possible to reach multiple gpus without the need of third party thread managers (MPI, intel TBB, etc.). No matter what the programmers use to create independant threads to reach the GPUs, they lose the ability to use OpenCL events for syncing calculations on independant GPUs.