The biggest problem in my opinion are inscriptions, because they exploit the witness discount and therefore bloat blocks up to 4MB, accelerating chain growth.
OP_RETURN data is less of a problem, since it is actually priced appropriately at 1 byte = 4 weight units.
So the ideal soft fork in my opinion would be to just get rid of the witness discount. Then there would be no incentive for inscriptions anymore, since they would be more expensive than OP_RETURN. And the block size limit would once again be 1MB.
The biggest problem in my opinion are inscriptions, because they exploit the witness discount and therefore bloat blocks up to 4MB, accelerating chain growth.
OP_RETURN data is less of a problem, since it is actually priced appropriately at 1 byte = 4 weight units.
So the ideal soft fork in my opinion would be to just get rid of the witness discount. Then there would be no incentive for inscriptions anymore, since they would be more expensive than OP_RETURN. And the block size limit would once again be 1MB.