Does too much normalisation create a problem?

All normalization does is set the highest peak in a file to a signal level you set. You've always been able to do this either as a percentage or a rather more useful dB value; it's never been set to any arbitrary value. If you are processing a 32-bit file, you can normalize to a really low value if you want, and then restore the signal to a more sensible level without loss, but if you do this with an integer-based (8 or 16-bit) signal and save the result before restoring it, you will raise the noise floor in your file irreversibly. So normalizing to a low value, apart from being pointless, isn't a good idea anyway.

Normalization doesn't alter anything about your file at all other than the overall level of it. It moves everything proportionally to the value you set, and the results will sound identical. It has two primary purposes: the first is that it sets any signal so that its peak reaches the level that all of the dynamics-based processing is designed to be referenced to - in other words, 0dB. So if you are going to use anything like a limiter or compressor, it means that when you set the threshold level it isn't arbitrary any more, which it would be if the audio was at some other level - you really wouldn't be able to tell what the true effect of the change you made had been. The second use of it concerns the final peak value you set a file to before encoding it to a lossy format. This is the one that is slightly more subtle. If, for instance, you are going to create an MP3 file from your final edited version, then it's best to normalise the signal to about -1 or even -2dB before doing so, as encoding has been known to put additional peaks into the output, and the results simply don't sound as 'clean' as they would with a slightly lower level signal to start with. With non-professional monitoring you are likely to hear a difference in the results of running at different levels, but this has nothing to do with the processing, and everything to do with distortion in your monitoring chain - the higher the level you run it at, the less clean it will sound. And that's not just the speakers; the amps running them will also distort more as well.

So, does any of this apply specifically to the hard limiter? Actually it does, even though it appears to have a maximum amplitude setting. If you consider applying it to a signal where the highest peak only reaches -10dB and have the input boost set to 6dB, then effectively the hard limiter will do nothing except add 6dB to the overall level, bringing it up to -4dB. You can try that for yourself - it's easy to prove. But if you normalize the signal to 0dB first, then having the max amplitude set to -0.1dB and the same 6dB boost will result in limiting the top 6dB of your signal, with the highest peak coming out at -0.1dB (allowing for the vagiaries of the look-ahead setting). So you will then have achieved your 6dB of hard limiting, with everything above -6dB squashed into zero dynamic range, which is what limiting is.

In other words, if you want to know what the hard limiter is doing, without normalizing the signal first, you simply won't have a clue. And as I said, that goes for all the dynamics processing effects.