Source engine concurrent client input serialization latency limit

The goal is to define the concurrent client input threshold at which a Source engine server's deterministic lock-step simulation begins to serialize inputs across a single tick and produce measurable latency spikes.

The server processes client commands sequentially per tick, so rapid simultaneous input from many clients can exceed the simulation step budget, particularly at lower tick rates such as 66 Hz. The exact overflow point is not officially documented, and netcode differences between Source 1 and Source 2 make the boundary uncertain without version-specific testing.

What concurrent input rate triggers serialization latency on a 66 Hz Source 1 server? Does Source 2's revised netcode raise that threshold, and by what margin? At what point does perceived lag persist past the next tick despite client-side lag compensation?