The only caveat I know of is that on a dual socket system you won't get twice as many lanes. 48 (or 64, depends on the configuration) of each are used for the interconnect to the other socket. Then of course the performance of any slot-to-slot traffic may depend on the different paths taken inside the interconnect, but that only matters for very high-bandwidth of low-latency applications.
It is true, the CPU-to-CPU interconnect fabric on both blue team and red team play into that, and it varies between generations and motherboards and chip sets. It gets very confusing quite quickly. Some server boards will let you dedicate PCIe slots to a specific CPU, others won't let you not dedicate certain slots to a CPU.
