Actually its the Tesla/Firestream cards that normally end up in those servers.
Id never heard of a server that uses say a Quadro/Firepro based cards since most of them end up in desktop workstation PCs for say a mechanical engineer working in solidworks on a project that involves 500+ parts.
And I always wonder how important those OpenCL benchmarks are especially when nVIDIA doesn't quite care about OpenCL because of CUDA.
I agree that these aren't server cards but are workstation cards. It doesn't mean they can't and won't work just fine in servers, though.
As far as workstations go, you don't generally see more than one high end card like this used, maybe two. If someone is going to set up a rack of these things it isn't going to be sitting on their desktop whirring away. While I'm sure 4 of these crammed onto an EATX motherboard will be loud and hot, 4 open air coolers aren't even an option.
As far as OpenCL goes, it's more important today than it was a year or two ago, and not as important as it will be in the future. Adobe and Autodesk are major forces in professional graphics. While CUDA is used it's to a limited degree, it appears to be losing it's foothold. HPC? Different story. AMD hasn't really made any inroads with OpenCL there. It's still early days though.