I'd strongly suggest getting a copy of Soares 'Grounding'. This is the _classic_ textbook on why we ground and why we ground in the way that we do. 'Grounding' is not easy, and the historical section will make it clear that grounding is not without controversy. In fact, at one point some of the early electrical safety organizations _prohibited_ grounding of electrical systems. (Check out the historical appendix of the book.)
Alas, the original author (Soares) is dead, and the book keeps being revised to match the current NEC, which means that it now reads as though it were written by committee (the same concepts repeated several times in ways that IMHO don't reinforce useful knowledge, having several examples in a row that look at different aspects of something, but using _different_ base examples when IMHO it would make more sense to look at the _same_ system form different viewpoints, etc.) But the text still covers the essential material, and is worth reading for this reason alone.
You don't need to consider jdadamo's failure condition to understand why the shared neutral for the cooktop is not desirable. The real problem is pointed to in your statement:
If the neutral and the ground are tied together at the main panel feeding the cook-top. Tieing the neutral and ground (from the cook-top) together at the junction box (using exhisting ground) right below the cook-top would be the same results.
You also said
And unless the neutral gets disconnected at the panel from the cook top, electricity will of course takes the easier path, not through the customer.
These are two common (and dangerous) misconceptions.
The first thing to understand is that electricity does _not_ simply 'take the easiest path to ground', or even the 'easiest path' in general. Electricity takes _all_ paths back to the _source_ of the current, in proportion to the ease of the path. If you arrange a system where the electricity has two paths back to the source, one through a thick copper wire, and one through a person, then _most_ of the electricity will follow the wire, and a small bit will go through the person. Make the difference in resistance large enough, and the current flowing through the person will not even be noticed.
Because electricity will take _all_ available paths, the grounding done in premises wiring systems is known as 'single point grounding'. All metal that _should_ be at ground potential is 'bonded' using equipment ground conductors and grounding electrode conductors. This makes sure that during normal operation all of this metal is at the same voltage, so that there isn't a shock risk. Then this entire 'grounded' system is connected to the neutral of the electrical supply at one, and only one, location. Because there is only a single point of connection, there is no closed path that the electricity can follow, and thus current does not normally flow in the ground system. That latter point is the crux of the issue.
The neutral is the 'grounded conductor'. It is insulated and _expected_ to carry current. The grounding system, on the other hand, includes essentially all other metal, and may or may not be insulated, but is _not_ expected to be carrying current. When you make _two_ connections between the neutral system and the grounding system, you change the story: now there are closed circuits in the grounding system, and current _will_ flow in the grounding system.
Now, if the metal frame of the range were bonded to the neutral conductor, rather than to an equipment grounding conductor with an isolated neutral, but the range itself were completely insulated from everything else, than I would agree, this would be exactly the same as the 'proper' connection back at the panel. But if the frame of the range has _any_ metallic contact with any other grounded metal, then the net result would be multipoint grounding and the introduction of current into the grounding system. If this current were flowing, for example, in the water pipes, then even without any sort of fault condition or failed conductor, you could have a shock hazard.
As iwire noted, previous versions of the NEC permitted the use of the _insulated_ neutral conductor to bond the frame of things like ranges, presumably because the total imbalance current of this load would be pretty small, and thus the amount of current potentially placed on the grounding system would be _very_ small. But it is interesting to note that this sort of shared neutral/ground was only permitted if the feed went back to the _main_ panel. If the feed went back to a subpanel, then the feed to the range had to be a four wire feed.
I'd also like to call attention to a situation where multi-point grounding is commonly seen, even though it is undesirable. In urban settings, it is quite common to see several houses, possibly an entire block, fed by a single transformer. Each house has their own ground to neutral bond in their main panel, and the city water supply pipe is used as one of the grounding electrodes. The city water pipe is a metallic system that connects several homes, and is clearly a 'parallel path' for neutral currents. In some cases, the neutral in the electrical system can fail, but the entire electrical system appears to function normally, since the neutral current is flowing through the pipes to an adjacent house, and then through the ground bond in that other house. Plumbers have been killed working on the plumbing pips because they've interrupted the neutral current, and experienced plumbers place jumper cables across sections of pipe that they are cutting apart. The next time you are in a home that has metal water pipes coming in from the street and a shared transformer, put a clamp meter around the pipe or the grounding electrode conductor to see what sort of problems multipoint grounding can cause.
[This message has been edited by winnie (edited 04-10-2005).]