I’d say you should always gain stage, but it’s particularly important early on in the signal chain. The main purpose is to avoid clipping, the 2nd purpose is to preserve dynamic range. If there’s distortion at some stage, it will persist in all subsequent stages. You have to watch the input levels. So if 3 vocals are mixed onto one bus, and the sum of these inputs reaches 0 dBFS, you’ll get clipping before the output fader on the vocal bus has a chance to act. The idea of 24 bit audio is to give you headroom that’s not supposed to be used (except as a backstop against clipping). A bit depth of 16 might force a trade-off between the risk of poor dynamic range and the risk of clipping.
Now, you should get warnings when clipping happens within Cubase, such as a channel strip label turning red and staying red. I’d say you’re good if that doesn’t happen. The deeper issues are earlier on, during tracking. If your audio interface clips, then the recording will me distorted and Cubase will not know it. You’d have to keep an eye on it, or count on some sort of limiter to prevent it. Some of this may depend on your vocal preamp, or the amp circuit in your audio interface.
Some people like the saturation effects of driving a signal into the red on analog gear, and we can get that with vintage limiter plugins. However, I prefer to get a clean recording and keep that option for later.
A common bit of advice is to use downward expansion to improve the dynamic range, and I did that until I got iZotope Rx software, which removes background noises of all sorts, but requires unrepressed noise to work properly.