This is some Sage code to check the counterexample posted by user1272680. I can't put this in a comment, so I am putting it as an answer, but the credit should go to user1272680.

```
g=Graph(21)
g.add_edges([(0,1),(0,2),(0,3),(0,4),(0,5),(0,6),(0,7),(0,8),(0,9),(0,10)])
g.add_edges([(1,2),(3,4),(5,6),(7,8),(9,10)])
g.add_edges([(1,11),(2,11),(3,12),(4,12),(5,13),(6,13),(7,14),(8,14),(9,15),(10,15)])
g.add_edges([(11,16),(12,17),(13,18),(14,19),(15,20)])
g.add_edges([(16,17),(17,18),(18,19),(19,20),(20,16)])
print g.chromatic_number()
for e in g.edges():
h = g.copy()
h.delete_edge(e)
print h.chromatic_number()
gd = g.to_directed()
[p for p in gd.all_simple_paths() if len(p) == gd.num_verts()]
```

First it constructs the graph, checks that it is 4-chromatic, runs through each edge and checks that the edge-deleted graph is 3-chromatic.

Finally it confirms that there are no hamilton paths - Sage apparently only has an "all paths" command for directed graphs, so I just turn the graph into a digraph where each edge becomes two opposite-directed arcs.