I think another issue with lockstep is that for it to work, each client needs all the input data from all the other clients to be able to simulate the next step, otherwise the simulations become out of sync. So how does the simulation handle both your input, and another clients input that comes in 100ms later? If you just pause until you get their input, your game is effectively pausing and running at 10 steps/second. For comparison Starcraft 1 ran at 42ms a step, or ~24 steps per second. You'd be playing in slow motion.
Games like Starcraft 1 would handle it by adding an artificial delay to your inputs (and all your opponent's inputs). Putting it into a buffer for future steps instead. If anyone remembers the latency option for "Extra high latency", this would increase the delay on inputs so that simulations did not have to pause while waiting for inputs from other clients. And then if any network fluctuations causes your input buffer of other clients to empty, you have to pause again.
In fighting games this is usually called delay-based netcode, and people hate it. It makes your inputs feel unresponsive, and in games that require fast reactions (like fighting games) gives you a smaller window to react.