There is a point there but those are also a couple of the most extreme examples. The eyes are delicate complicated organs very isolated from the rest of the body and changing blood characteristics requires sterilizing and replacing your bone marrow which is crazy complex and dangerous. There are many more mundane targets for gene therapy.
Also one of the benefits of cheap to produce unique gene therapies is once you figure out one treatment, similar treatments for slightly different genetic anomalies affecting the same tissues will become much easier to apply. In other words once you have a delivery mechanism down targeting something specific, the gene payload can be swapped at much lower incremental cost.
That second point is what the LUMEOS trial demonstrates is not true: LUMEOS tested a AAV2 vector delivered inside the retina by a surgeon (just like Luxterna) yet instead of 80% of patients showing improvements only 40% did(1). What went wrong? What is the difference? That's going to take more research, which makes the whole thing expensive and time consuming and risky.
We are not yet at the point where we can effectively just substitute genes and everything works in vivo, and until we get to that point every fix requires its own Phase III trial, and that (both the costs and the chance of complete failure) is what drives the R&D expense through the roof, even if a single infusion could fix the problem, unlike the other examples I cited.
1: Existing gene editing platforms don't work well for large genes: there is an effective base pair limit for all existing techniques in vivo. Most Stargardt's genes, to pick another similar to Luxterna as an example, are above that limit, and will require new techniques not yet approved in vivo.
Also one of the benefits of cheap to produce unique gene therapies is once you figure out one treatment, similar treatments for slightly different genetic anomalies affecting the same tissues will become much easier to apply. In other words once you have a delivery mechanism down targeting something specific, the gene payload can be swapped at much lower incremental cost.