The 2024 EOS R1 and 2024 EOS R5 Mark II do not use "Sensor Shift" (IBIS High Res) to create extra pixels. Instead they use a Neural Network Upscaling tool. This tool uses AI to predict and add pixels to a JPEG or HEIF file after it is taken. You are correct that this does not add "real" optical data. Instead it estimates detail based on deep-learning patterns.
Heat: Constant sensor movement for high-res shots creates heat which conflicts with the thermal demands of 6K/8K video and high-speed bursts.
I don't think heat is an issue, given there's plenty of things that I can do on my R5 that generate a lot of heat; cooling is vastly improved on the R1 and R5 II and this feature is not used in rapid-fire scenarios; rather, it's used for landscapes and situations where images are carefully planned.
The AI upscaling is adding detail and artifacts that are not there. It's not the same, and lens artifacts can creep in and increase the margin of error. The goal is to capture more detail, not AP upscale a jpeg in-camera. I don't see any technical reason why this can't be added to the newer EOS IBIS-equipped bodies. AI slop is not the way.
One huge use case I have is for high resolution macro images, where this will get me far more detail. AI fabricated slop is not going to help me in this use case.
As with all these things, I am fascinated from a product management perspective as to why this feature wasn't made available on the R5 II at least. It's perhaps the one and only reason why I have kept my R5 and not replaced it with the R5 II.
The Sigma 35mm f/1.4 DG DN Art uses a large optical formula to correct distortion physically. The Canon RF 35mm f/1.4L VCM is designed to be a "Hybrid" lens. It prioritizes the Voice Coil Motor (VCM) for near-silent instant focus in video. To keep the lens small and the AF fast Canon uses digital correction to fix the extreme barrel distortion.
As you noted stretching these pixels in software causes a loss of raw detail in the corners.
On the 2022 EOS R7 this is less noticeable because the "crop" sensor ignores the lens corners. On the R1 or R5 II the "software stretch" is visible in large prints.
It's not just the software stretch and loss of resolution, but the way this interacts with the noise pattern. Noise from sensor gain at high ISO should be uniform and consistent across the whole image, however once we change the geometry, we will be squeezing in and stretching out the noise, creating artifacts that look like a lattice, moire and so on. Furthermore, this conflicts with noise reduction principles that assume a consistent noise pattern across the image. This is further compounded b stronger vignetting on some of the RF lenses; I don't mind the vignetting look sometimes, but I don't often get to choose anymore due to these two features conflicting; that's around two stops on the RF 35mm f/1.4L VCM, so you've now got an image with ISO 51200 exposure in the corners of an ISO 12800 image, which in darker situations, results in near-unusable images if any corrections are made - and by my testing, the vignetting falloff occurs closer to the centre of the frame than it does on the EF variant. So whilst the t-stop is measured from the centre of the image in both cases, we ned up a lower potential average t-stop on the newer lenses (using this to account for transmission loss, which impacts low light performance at the same aperture).
Re: the Sigma - the new Sigma is shorter and lighter than the RF 35mm f/1.4L VCM. I don't believe that the optical formula is any larger on the mark II vs the RF 35L VCM. Furthermore, the testing and reviews so far don't seem to suggest any significant penalty for autofocus. The only aspect I don't understand is whether the VCM motor is physically larger and is taking up space that is forcing the optical elements to be of a smaller diameter. I appreciate that in video situations, the VCM lens distortion is less relevant, but still an issue for high gain/high ISO situations due to the aforementioned issues. There is no excuse to have a 35 with that much barrel distortion in 2026. This is a solved problem that has been un-solved.
It's a snowball effect. I was jumping for joy when the EF 35L II was announced, as optical distortion on the first-generation EF 35L was a bugbear of mine. I can only hope that this is swiftly addressed. I wouldn't be so frustrated about being left high-and-dry if third party RF lenses were available.