I used up all my Claude tokens for today on this issue. Every change to the source code led to major differences. Basically, esi is assumed to be 0 by the msvc optimizer and uses this to its advantage to produce a byte shorter (faster?) code. I tried combinations of optimize pragma's, including options "gy" produced the 91.53%.

Does the StructResolver shortcoming resolve itself (no pun intended) when DAT_SoundEffectsHelperData1 becomes reimplemented?

Ok, I guess then it does not make sense to use more tokens here unless you want to try one or two iterations on the raw Ghidra export, if you have not already done so. But this is up to you.
One fear I have is that it might actually be related to the position of the function in the original: That it just decided to use another opcode to the certain jumps aligned. But this still seems unlikely.

I actually had to notice something: The struct resolvers do not work together with reccmp. I was unable to properly annotate the static created by the template, even with tricks. Their repo reads like normal statics should work, but I have no idea how to implement a template static.

This might be something to ask them, I guess?

Yes! They have a Discord server too. disinvite is the most important developer I think currently.

Can you provide a code snippet of how you expected things to work?

I can provide a reduced version that I tried, which seems to not have worked.
I tried to use the "instance" Wrapper directly:

struct StructResolver {
    template <typename T, int gameAddress> struct Instance;
}

// Try 1
// GLOBAL: TARGET 0x00DF37F0
template <typename T> struct StructResolver::Instance<T, 0x00DF37F0> {
    // Try 2
    // GLOBAL: TARGET 0x00DF37F0
    static T instance;
};
// Try 3
// GLOBAL: TARGET 0x00DF37F0
template <typename T> T StructResolver::Instance<T, 0x00DF37F0>::instance;

// Try 4
// GLOBAL: TARGET 0x00DF37F0
StructResolver::Instance<ActualType, 0x00DF37F0> testInstance;

Non of this seems to work. (And it is a bit simpler as our actual resolver, which currently hiddes the instance and only exposes a ptr.)
I do not really think any of these truly express the template instance global. Even 4 would be wrong, since the static does not truly belong to the wrapping struct.

Which also means I actually do not really have an idea. Maybe you can attach it to the specific template instance somehow and just define it in relation to the template? No idea. But at least the reccmp output properly shows the instance, it can just not attach the annotation to it.