Are there cases that should be referred to a center with a MR Linac for RT?

There likely are cases that would be best treated on an MR Linac, though I suspect they are relatively few and far between. The categories noted in the question stem are certainly considerations, however, I feel that only a subset of these patients would gain true advantages. MR Linac offers the potential for better motion management as well as improvements in soft-tissue imaging and artifact. 

Most of the ultracentral lung cases I have treated (with a standard linac based approach using ABC gating as tolerated, otherwise abdominal compression w 4DCT) do not have significant motion (i.e., more than 3-5 mm at tumor/PBT interface), and as such, likely do not stand to gain substantially from MR Linac based gating. Perhaps the real-time imaging of MR based approaches could make one consider a smaller PTV margin, however, triggered imaging with a standard Linac verified to PBT or tumor is also quite reasonable to confirm consistent alignment during treatment while maintaining full advantages of fine MLC conventional IMRT/SBRT treatment. Patients in whom motion cannot be successfully managed would likely benefit from MR Linac based approaches.  

Similarly in the re-treatment category, I would consider primarily only re-treatment cases in which motion drives dose to exceed the constraints of a serial OAR as most appropriate for referral to an MR based Linac. 

Pancreas SBRT (but not conventionally fractionated pancreas radiation) is likely the most appropriate consideration for MR Linac referral, as it checks all three boxes for potential advantages: 

1. Potential for adaptive treatment planning to improve therapeutic ratio given tumor in close proximity to serial OAR's which frequently change position (and a relatively radio-resistant tumor which we often struggle to control with current approaches)
2. Potential for significant intrafraction motion which could be managed by MR based gating (though also can be managed with ABC and fiducials on standard linac)
3. Placement of high radiation dose gradients in an area where CBCT frequently has significant imaging artifacts near tumor/sensitive serial OAR interfaces. 

Even with these, potential advantages of MR Linac in pancreas SBRT comparing data when available will be critical to determining whether MR based approaches truly realize these advantages in a way that conventional SBRT cannot. As a disclaimer at CCF, we do not currently have an MR based Linac available, though still feel comfortable being aggressive with the dose for pancreas SBRT on a conventional Linac. Utilizing ABC (or if poorly tolerated surface based) gating approaches verified by fiducials and triggered imaging we feel comfortable with motion management. Tumor volume more than 7 mm from stomach/small bowel/duodenum is escalated to a goal of 50+ Gy/5 fractions in treatment planning (while maintaining hard constraints on those serial OAR's at the expense of tumor coverage if needed) with remaining tumor in closer proximity to OAR's prescribed to 30 Gy/5 fx. The 25, 30, and 40 Gy IDLs are turned into a structure to verify location relative to OAR's at the time of delivery. While admittedly, there is some uncertainty at the time of treatment due to imaging artifact, toxicity to date with this approach has been generally favorable (and very few treatment fractions are judged to be untreatable at time of SBRT). Fractionated approaches such as MSKCC 15 fraction approach may also mitigate some of these concerns.  

MR based Linac does offer some theoretical advantages over conventional approaches, however, the specific scenarios in which clinical benefit is likely to be realized are likely quite narrow, and much like other advanced technologies, additional data is needed to demonstrate the value of the technology for universal adoption. In a world where healthcare resources are not unlimited, it is important to ensure we allocate spending in areas with the greatest potential advantages - and many disparities in our system remain unaddressed.