As a general principle, keeping it simple is a good one. Another good general principle is optionality. In this case they’re in tension.

Yes, OP_CAT opens the design space—as did SegWit and Taproot to an even greater degree. Having more programmability does mean that some people may use it in ways we don’t like. The question is what harm can they do by it?

To me, it’s crucial that we keep the cost of running a node down. More validating nodes makes a stronger, decentralized network.

It’s also crucial that we keep mining decentralized. Having participatory mining and node operation are both important.

So if OP_CAT would lead to miner centralization, or substantially increase the cost of running a node, those would be compelling reasons to oppose it, IMO. I haven’t heard anyone make this argument though.

The argument I’ve heard so far is as you say “it’s harder to predict what can be done with it”. Yes, and also that affords creativity in the face of our adversaries.