Based on your problems with low-bandwidth/high-iops reads & writes on multiple PCIe raid cards, it really does sound like you are dealing with a PCI bus contention issue.
Could you try running this tool while a benchmark is in progres:
DFI Latency Checker
While my SSD was connected to the mobo's built-in SATA II controller, I tried the DFC Latency Checker while and ran AS SSD Benchmark. There were no issues!! The maximum latency that it reached was about 126us which is in the green zone!.