This feels like some seriously missed static analysis optimization though - why do I need to specify this? There's vanishingly few cases the answer to "looks like you don't modify this ever, make it readonly?" where the answer is not yes.
Probably because reflection & compilation+activation of types existing outside the original compilation context is possible at runtime.
I much prefer to having a language/runtime that can handle extremely complex problem scenarios out of the box (with the need for occasional tuning), rather than the other way around. It is not very often that I find myself screwing around with performance-intensive code. Usually, a few very important things are addressed at the lowest levels and then you put all your nice abstractions back on top for daily driving.
