Thats a specific problem you will also find in other synthesis approaches like additive synthesis, where its not obvious how to create an expressive control interface. For that reason you will find a subtractive-additive approach in all modern additive synthesizers (razor, harmor, pigment etc.) which have function based macro controls instead of individually setting amplitudes and phases of partials. The same is true for these DX7 like operators. When setting numbers by hand you will very unlikely have an expressive instrument. One way of dealing with that is to use the MLPRegressor from the FluCoMa toolkit to interpolate between different presets, which could be created randomly and then you mostly care about parameter ranges and not so much about specific parameter values, which arent self explanatory anyway because of their non-linear nature. You will find alot of literature where this behaviour is even described as favourable over linear a to b mapping, lose control gain influence. For example in the PhD thesis by Tom Mudd Nonlinear Dynamics In Musical Interactions
You could start from both ends of the spectrum by all operators having their own operator-specific parameters and all operators sharing the same parameters and then look how to get and shape the sounds you are interested in a way which suits you best.