I read this yesterday and found it consistent with thoughts I was having about where the final maturity of development will end on both systems and how they'll compare.
Like you, I was thinking that the CU count they've got in the XsX isn't going to be a linear gain, no matter how well they parallelize the workload situation with advance software tooling.
Cache availability per shader array seems to be less per CU in the XsX, even if they are limiting the arrays to 9 active CUs to match the PS5 setup in these early games - just at a clockspeed of almost 400mhz less.
Looking at the faceoff data so far it does feel like a comparison of a 36CU GPU at 1.8GHz versus a 36CU GPU at 2.2Ghz with cache scrubbers - where reducing the frame-time, by upping the frame-rate, pushes too hard on the GPU with less cache bandwidth per frame, and less clock ticks per frame, that then results in needing to reduce LODs and resolution to stay within the data & processing bandwidth as seen in that headlight comparison picture on Dirt5 in performance mode.
Putting aside the major changes that we might expect to see from more use of async and Cerny's geometry engine on the PS5, I'm at a loss as to how the XsX can effectively utilise all of the CUs to its advantage with ICE team level solutions to get to PS5 level day 1 performance/fidelity without XsX needing to target a lower resolution that offsets its deficiency in rasterization disproportionally beyond the clockspeed to freeup cache bandwidth for the extra CUs to contribute evenly with a fully parallelized workload.
I've seen comparison comments about the PS3 and XsX situation and how it will improve, but where the PS3 SPUs are self contained with embedded memory local stores - and could DMA individually or any workgroup combination independent of the PPU once setup and kicked off - the extra XsX CUs within an array will contend for shared cache on every read and write, and won't even have the half-way-house situation of the Tempest Engine.
Like you said, a DLSS solution is probably the intended use case, although I wonder if tessellation could also save cache bandwidth to make up the current difference., even if the DLSS/tessellation might need to be more aggressive coming from lower LOD levels and lower native resolution for high frame-rate modes I expect the XsX with fully advance tool options to exceed the IQ of the Dirt5 120hz mode on the PS5.
Your comment seemed like good food for thought, anyway and even had me wondering if the tools narratives could both be true. Xbox tools to port games are ahead of schedule and not coming in hot, but still far away from the ICE team level solutions they need to match PS5 day1 results with Sony's tools that are coming in hot.