There are a number of more complex alternatives under discussion which involve more elastic limits (instead of the hard limit which currently exists), automatically adjusted limits based on some metric in the blockchain, or some combination of both.
More complex alternatives mean more complex code and more potential for bugs. They certainly could and should be tested on testnet.
With the "just raise the hard limit" option (and even with some of the other alternatives), most metrics you could glean from a testnet test aren't very interesting and are already easy to calculate. E.g., it's easy to understand what the worst case bandwidth, CPU, and disk space requirements are, and it's not too difficult to simulate (without testnet) how added block propagation time will (or will not) affect miner efficiency.
It's the external effects which are causing most of the controversy, e.g.:
- Will the additional requirements drive more people away from running full nodes, increasing centralization?
- Will Moore's Law, Nielsen's Law, and the current Kryder Rate continue into the future, and will they help to improve decentralization despite larger blocks?
- Will a "don't do anything until it's an emergency" approach drive people away from Bitcoin when tx confirmation times begin to significantly increase (and Bitcoin Core nodes begin crashing)? Or will tx space scarcity simply cause tx fees to gradually increase (which will also drive users away if they increase too much, increasing centralization)?
- What effect will the blocksize limit have in the future (e.g. on tx fees) as the block subsidy decreases?
I'm not trying to list all of the controversial questions/issues (I doubt I even know them), only to demonstrate that there are far more which can't be answered by testnet than can.