Simulating films is an interesting challenge and sometime in the distant future it might be something I'll try to develop software in.
The philosophical question "why bother?" I'd answer like this: think of photographers that want to learn shoot film, or those that shoot both film and digital, and want the look to match. So you can work with light meters etc and learn film shooting technique without wasting any film frames. For this to work the film simulation must behave quite accurately. Personally I don't really care what the reason is, I find film simulation to be an interesting technical challenge.
But let's first answer the question why the look is so different. Many of the simulations are simply by eye manual hand-tunings for some specific picture, the HALD package for rawtherapee is that AFAIK. Others like DxO (and RPP I assume) are based on test charts which I think is more reliable. Still it will probably depend quite a bit on which light the matching was made, which likely is something similar to D50. How colors are modulated when you change light is probably quite different from film to camera, which is something I'd like to experiment if making film simulation software. I think it may be necessary to make a multi-illuminant simulation to match colors well over a wider range.
Then we have how film contrast is simulated. Many don't simulate contrast at all and let you freely set the contrast in the software. This is not so good if it comes to simulating film behavior, as the contrast is an integral part of the look. Velvia has very strong contrast, looks "black crushed" in many scenarios, so to learn shooting well with Velvia one have to be an expert on its contrast.
Then we have over-exposure simulation. As far as I know no current film simulator simulates film over-exposure in a good way. Some aspects are really hard to simulate, such as the center of the sun becomes yellow on the film, that would require raw reconstruction in the raw converter to be able to simulate at all (as a digital camera clips straight off). But assuming we have no raw clipping and then push the exposure slider, no current film simulation will simulate the film's behavior but instead used their standard "digital" over-exposure handling.
I don't know this, but I would suspect to simulate some of the "color quirks" of real Velvia you'd need to have a really aggressive LUT that could cause broken gradients, and then the tradeoff between smoothness and simulation accuracy will probably affect the result quite heavily. Commercial profiles always prioritize smoothness, robustness easy of use etc over accuracy so I wouldn't trust Fuji's simulation to be most accurate.