Normalising doesn't remove any dynamics. The relationship between the loudest and softest bits in your tracks stays exactly the same after normalisation--all it does is move the loudest bit to the level you have pre-determined with everything else changing by the same amount.
It's compression that can mess with your dynamics.
As such, normalising is simply a way of moving the overall level of the whole track a bit. It doesn't actually change your mix but it can make the mixing process easier by putting the levels in a convenient range.
However, there's a potential drawback of normalising if your DAW is stuck with integer operations rather than the superior floating point.
If you had an integer file - let's say a 16-bit one - and decided to reduce the amplitude of it by 36dB in your session, the session would ask you whether you wanted to save the file. If you said yes, then you'd have a 16-bit file with a maximum possible amplitude (and therefore signal to noise ratio) of 60dB (you can have up to 96dB of dynamic range in a 16-bit file). But, that amplitude, having been reduced by your DAW, will result in your signal not using up any more than 10 of the 16 bits available.
The trouble comes when you re-open the session and decide that you want that file back at its original amplitude. You've now got a rough-sounding 10-bit file masquerading as a 16-bit one...
Okay, that's quite an extreme example, but the basic situation is the same with all integer files. It's less of a problem with 24-bit ones, because you have a lot more bits to play with, and even if it reduced to 16 bits it would sound fine. The real issues with integer files from this point of view is what the reduced bit depth does to quiet things like ambient sounds, and natural reverberation - it screws those up a treat. The absolute joy of the Floating Point system is that all the actual signal data remains intact (although in a normalised form). It's just the scaling that is altered - so reducing the amplitude, saving, and then increasing it again results in a file that is identical to the original; no losses at all, as the bit depth has now become insignificant.
So, with integer operations, there are theoretical downsides to normalising and saving a file (though, as I say, at 24 bit depth it's not really an issue) but, as long as your DAW supports 32 bit floating point working, then normalising is 100% lossless and 100% reversible.