I would love to be able to add this as a comment as this answer isn't really in the spirt of puzzles but I lack the reputation.
I wrote some python code to brute force solving the problem.
From this I found,
If we let $N$ be the starting number of chips for each player.
For $N=1,2,3$ Bob and Charlie will each win with probability $\frac{1}{2}$
For $N=4$ Bob and Charlie win with probability $\frac{7}{16}$ and $\frac{9}{16}$ respectively.
For $N=5$ Bob and Charlie win with probability $\frac{9}{20}$ and $\frac{11}{20}$ respectively.
These results are weird enough that I used an exact fraction module to check it wasn't a floating point error. Of course it could just be that I wrote bad code (this is somewhat likely) but if not then I don't know why this happens.
Just to make this a proper answer
Charlie is most likely to win.
Update: Here is the code I used.
import numpy,itertools,fractions
def partitions(n, b):
masks = numpy.identity(b, dtype=int)
for c in itertools.combinations_with_replacement(masks, n):
yield sum(c)
def Probs(S,a,k):
if k>n*N:
print('panic')
if len(S)==1:
return [fractions.Fraction(1)]
Best=[]
Max=0
for U in partitions(k, len(S)-1):
V=list(U)
V.insert(a,-k)
R=[a_i - b_i for a_i, b_i in zip(S, V)]
if any(t < 0 for t in R):
continue
L=[i for i in R if i != 0]
b=a+1
b-=R[:a].count(0)
if b==len(L):
b=0
vals=Probs(L,b,k+1)
newvals=[]
t=0
for i in R:
if i==0:
newvals.append(fractions.Fraction())
else:
newvals.append(vals[t])
t+=1
if newvals[a]>Max:
Max=newvals[a]
Best=[newvals]
elif newvals[a]==Max:
Best.append(newvals)
P=[sum(i)/len(Best) for i in zip(*Best)]
return P
N=5
n=3
I=[N for x in range(0,n)]
print(Probs(I,0,1))
Update 2:
Here is my initial understanding of why this is happening. The analysis given by @SQLnoob is correct in determining that on that branch Alice will choose randomly, however it makes the mistake of assuming that Alice cares about their actions earlier in the game. Under the assumption that Alice can never win (among the traversed paths) all of Alice's choices are uniformly random as they are all equally good. This means that it is possible that some earlier paths get chosen which are not equal for Bob and Charlie. This is why there is a discrepancy. Presumably most paths are equal for Bob and Charlie, which is why the probabilities are so close to $\frac{1}{2}$.