To sum things up:
As Nick said - in a _true_ 32bit-floating-point environment, the only place where you should take care fory any overloads is the point where the digital signal gets converted to a fixed-comma format, either for analogue output, or for digital connection to another piece of gear via conventional" S/PDIF-, AES/EBU- or ADAT-cables. Everything what is above 0 dBFS (read: "Full Scale") will distort the signal to some extent. A brickwall limiter as last instance sould take care for this, keeping the occasional overload just below 0.
As long as you send a signal from one (virtual) channel to another, the 0 db-mark can be considered to be a guiding line, but due to the enormous internal headroom of modern 32bit DAWs (high above +1000 dB!) there's no need to worry. The same is true for proprietary connections like FXTeleport and the like. The only problem you _may_ run into are plugins that don't do their calculations in true floating point math. Therefor, as a general rule, it is always a good idea to keep the gainstaging under control.
Of course, there are other concepts of DAWs like ProTools, which use no floating-pont engine. On theses systems, each red light has to be taken serious.
It is good to keep in mind that the "0 dB"-mark means totally different things in the digital and the analogue world. While there is actually no "life" above 0 dBFs in the digital domain, a 0 dB-mark in the analogue world is just a convention. It depends a lot on the context what actual level is reached when you hit these zeroes.
HTH,