My experience is based on LS3/5as.
I had an original Rogers 11ohm pair with it's fairly complicated crossover.
When the Cicable external crossovers came out for them I bought them and removed the internal crossovers - I also replaced the cabinets with thinner walled ones (9mm instead of 12mm like in the original Kingswood Warren built prototypes) so that might muddy the picture slightly.
The Cicables were designed by Derek Hughes, who used to run Spendor and learnt his craft from his father (designer of the BC-1) and at the BBC, and has since designed all of Graham Audios and Stirling Broadcasts speakers and a passive speaker system in the Royal Opera House in London, which is very unusual as such systems are normally active.
The crossovers use expensive inductors and caps from Mundorf and at points reduce the THD of the LS3/5a by about 40%.
The difference is stark - much more transparent with more detail and much smoother.
Derek apparently doesn't just implement the standard type of crossover curves, he considers the driver characteristics more - there are interviews with him on YT where he mentions this more.
Derek designed the Stiring Broadcast V2 LS3/5a when Kef stopped making the LS3/5a drive units, which use modern drivers from SEAS and Scanspeak.
The crossovers for these are less complicated than the originals. They sound good but not as good as the Cicable crossovers which is not surprising considering the price point they were designed for.
Later they produced the V3 version which is basically the same crossover but with the iron-cored inductors replaced with large air-cored ones. These are more transparent and detailed, closer to the Cicables, but there is a 'shizzle' to the upper end, maybe exposing some of the THD in the original V2 crossover and it's more basic components.
Now there is a V3.2 version which has a more boutique crossover which is handbuilt by Derek, or his daughter who works at Audionote). Components are Jantzen air-cored inductors and Crosscap capacitors. Again another step up from the V3.
I also have some Xtracable external crossovers, also by Derek H, for my V2s and they are again slightly better than the v3.2s - also using expensive Mundorf components.
This progression in sound quality is whilst using the same cabinets and drivers, so clearly isolated.
Lastly they had a bass extender called the AB-2 which used a band pass conenction to the LS3/5as with a sinlge 15mH iron cored inductor with a 220uF electrolytic and 6.8uF bypass polyprop capacitor per channel.
With the v3 or v3.2 connected via the band pass there was a loss of audio quality, often seen with a reduction in the soundstage.
Replacing theses AB-2 crossovers with a Jantzen air-cored inductor and Jantzen crosscap or Mundorf Evo Oil caps completely stops this loss of audio quality, most easily seen in a retention of the soundstage. The bass is also very slightly better but limited by the nature of the AB-2 bass extender.
Commercial speakers have crossovers that are very much price constrained compared to what Danny does. The v3.2s that Stirling are doing are using crossovers that are quite expensive and not commercially sensible, but they are closing down and it is only because of that and that there is an LS3/5a enthusiast building them (me) that they exist. If you were to buy a commercial LS3/5a with a crossover that expensive you would be looking at several times their cost.
Same with those AB-2 crossovers, about £480 in parts alone. Worth it to me and a few LS3/5a nutters that have signed up for them, but not commercially viable.