All of the GM and XG Lite and most of the XG and GS voices are single element (single sample) sounds in order to conserve polyphony, mainly for song sequencing. If you have a 64 voice keyboard and are playing one voice that is single element, then you have 64 voice effective polyphony, but if you are playing a two element sound, then the effective polyphony drops to 32. You can see, that if you layer a two element piano voice and a two element string voice, your effective polyphony drops to 16. At this rate, with midi backing tracks for a full orchestra, you can very quickly run out of available polyphony, so the above systems stuck mostly to single element sounds to avoid this. Many of the "native" or "front panel" voices of your keyboard are multi-element sounds developed to sound richer and more realistic than the older single element sounds, but even the single element "native" voices will still sound better because they used the latest technology available when your keyboard was designed, while GM, GS, XG, and XG Lite voices are a throwback of at least a decade or two. They were included merely to keep your keyboard backward compatible with standard midi files designed to use those voice systems.
There is probably not much you can do to enhance your XG Lite voices. I would save them for sequencing midi backing tracks (and then only if you run into polyphony problems) and use the "native" voices for your lead sounds.
I can't help but wonder if this is, at least in part, inherent in the E423. My PSR-3000 has GM, GS, and XG voice sets, and I have a lot of midi files that use them. Even though they don't have the richness and realism of the 3k's "native" voices, they really don't sound all that bad.
Hope this helps.