My take on lots of this discussion in no particular order:
There's nothing inherently evil or bad about normalisation in itself. It has no more effect on your signal than pushing a fader up or down--but with the extra bonus that you can specify the maximum level you want things to go to--so even a single high level transient will not go into clipping.
Normalisation is ALWAYS reversable (until the final fix of course) even if you don't have an undo function. Since it's not making a permanent change to your recording, simply normalising downwards returns your recording to exactly what it was before.
Possible exceptions to the above are when you mix partly in the box and partly with external hardware. Depending on how your outputs are calibrated, normalising to nearly 0dB(FS) MIGHT result in your external hardware receiving a signal that overloads its input because it's at around +18dB(VU). However, since most commercial recordings are right around the zero mark, your sound card output is likely not to be giving an analogue +18 even when the digital levels are that high. You have to know your gear. Similarly, I've heard that some plugins (especially those that emulate analogue distortion) sound bad at higher levels. I've not personally encountered and plug in that has this problem...but I use very few 3rd party plugins.
Having said all that, normalising everything to LOOK the same is a fairly meaningless exercise. Raising the levels won't help you see brief transients, it'll just put those transients that much closer to clipping. If you really need to look for problem transients, zoom in on the horizontal scale and examine your file almost on a sample by sample basis. And, if you DO have one or two transients seriously out of line with the rest of the track, this is the time to pull out the compresser/limiter. Since (as has been said endlessly in this topic) you don't even hear level changes on transients under 200ms, limiting those slightly won't be audible but will give you a safety cushion. As always, how it sounds is most important with the metering being the most reliable tool to check levels against clipping. The graphic display is more useful in editing than mixing.
However, even that should be largely unnecessary. If you're working at levels where you have to be concerned about transients clipping then you're probably recording too hot.
The other disadvantage of normalising upwards is that, in the mix, you just have to pull down all the levels anyway. As you add tracks together, each one brings you a few dB closer to clipping on the final mix. Indeed, this is yet another reason to keep your levels reasonable when tracking. Indeed, I've been known to normalise downwards just to keep my faders or volume envelopes on certain tracks in the more linear section of things.
Finally, you can eliminate a lot of these issues by working in 32 bit floating point format.
Anyhow, my two cents worth on lots of things.