I don't see any problems there. The computer always needs a certain amount of time to process audio properly. No matter how fast the CPUs of the computers are, fewer buffers than 32 samples @single speed is hardly possible, and I have never seen this implemented before. RME manages very low latency across the entire product range, no matter whether USB, PCI, PCIe, or Thunderbolt.
And with these cards even with a massive number of channels.
But we are only talking about a few samples of latency for the internal forwarding from/to PCIe ports or MADI ports on the "backplane / the FPGA" on such a card. The modern converters on the connected devices also only require a few samples for the conversion (AD/DA) of audio.
This is more or less the same for all current cards, regardless of whether the converters are directly on board or connected via ADAT or MADI.
With LAN/AVB I have heard that the buffers can still be adjusted a little, depending on how large the network is, but here too I would not expect any major leaps in the end (regardless of the speed of USB, PCIe or the LAN). Ultimately, the audio data cannot flow any faster than the sample rate.
So I would say it doesn't matter what you take, important is the feature set, that makes most sense for your setup and workflow. Number, type of ports, etc ..
For you, the optimizing driver of the HDSPe MADI FX also could make sense, not sure, whether such driver optimizations are planned for the new cards.
BR Ramses - UFX III, 12Mic, XTC, ADI-2 Pro FS R BE, RayDAT, X10SRi-F, E5-1680v4, Win10Pro22H2, Cub13