When you use the randomize button to generate your starting gold (5d4 for fighter for example), a distribution like this is produced:
(500 Samples)

DNDBEYOND
{5: 34,   
 6: 30,   
 7: 49,   
 8: 31,   
 9: 25,   
 10: 32,  
 11: 27,  
 12: 18,  
 13: 35,  
 14: 35,  
 15: 24,  
 16: 31,  
 17: 25,  
 18: 33,  
 19: 37,  
 20: 34} 

 The expected distribution for 500 5d4 rolls is something like this.

EXPECTED:
{6: 3,    
 7: 6,    
 8: 19,   
 9: 33,   
 10: 51,  
 11: 59,  
 12: 72,  
 13: 87,  
 14: 68,  
 15: 54,  
 16: 28,  
 17: 13,  
 18: 5,   
 19: 2}

Thats a pretty important feature and should work. The normal dice roll function seems to work as intendet.

If that's right then it's probably just picking a random number between the minimum possible and maximum possible - instead of doing each dice individually.

Yes that is definitely what is happening.

You can easily test this by just clicking a few times.

The odds of getting a 5 or 20 with 5d4 are really really low.

When you just click a few times you will get quite a few 5, 6, & 19, 20 results which should be much much rarer.

This is still not fixed.

And it probably never will be. ¯\_(ツ)_/¯

Not much of a difference to be worth changing the coding, it's more system drag because it has to do more calculations, and hardly anyone uses random wealth rolls instead of just using starting equipment. And among those who do use it, only a small amount are going to give enough of a crap to "need" it changed.

And it's gonna be changed when 1D&D comes along next year so what's the point adding changes now, just to make more later?