AMD Opteron CPUs are marketed for servers, and FX CPUs for desktops. However, both support ECC, and while only Opterons support registered RAM, the issue is unrelated to reliability.
So, what makes Opterons more suitable for servers? Are they more resilient to single-event upsets in the CPU itself?
The difference lies in the intended application of the CPU... Granted, I come from a world where I don't assemble servers from discrete components, but rather purchase purpose-built equipment (from a manufacturer... with a warranty...) If I were in the market for an AMD-based server system, the candidates would all use server-specific Opteron CPUs... The FX processor line isn't even on the radar by the large server vendors.
To your question, you ask about reliability... CPUs don't fail often, so I doubt that there's a reliability difference when the right operational conditions are met (airflow, heatsinks, power, etc.). The real differences between the processor lines are likely cache, thermal properties, expansion, power consumption, duty-cycle and I/O capabilities.
With server's, especially for blade systems and the 1U pizza boxes, an important constraint is space, which limits the size of heat sink that can be added to a CPU. That in turn limits the amount of heat dissipation that can be achieved.
That with a requirement that they run can 24x7 for years on end is one of the reasons you see lower powered CPU's in servers as well.
A consumer grade CPU, especially one geared towards gamers, can sport an enormous heat sink and cooling assembly, simply because there's more space in the case. The top of the line AMD FX CPU's are rated at 220W...
Second server grade CPU's typically get deployed in two, four or eight socket configurations, which is not a common consumer requirement either.
And that doesn't even touch on instruction sets within the CPU that may be more relevant for business purposes or home entertainment.
