What are the conditions for using "Smith et al., unpublished data" in journal articles?

The fact that the data behind a claim is unpublished and unavailable obviously makes it hard to establish the veracity of the factual claim purportedly supported by that data. However, it does not necessarily mean that you can't confirm that a factual claim was made, and in some cases that may be relevant information for your paper. The example you give is:

"It has been suggested this figure may be in excess of 800 gazillion dollars..."

In some cases, the fact that a particular estimate of an unknown figure was made might be relevant information for a paper, even if only to give some context for the range of estimates people have come up with for an unknown figure. That is a weaker claim than if you remove the middle-man and just say that:

"This figure may be in excess of 800 gazillion dollars..."

In the former case, the claim is just saying that someone else made an estimate of an unknown thing; in the latter case, the claim is the estimate of the unknown thing. To confirm the first statement, you need only show that there is a work that makes this estimate; access to the data is not required. However, to confirm the second statement the actual data backing the estimate becomes important.

Now, it is probably not good practice to make the second claim above and then cite unpublished data, because it will leave your reader skeptical of the estimate, and they will blame you if it turns out to be baloney. However, if you just report to them that someone else has made an estimate, but the data purportedly supporting the estimate is not available, you are not weighing in on whether or not it is a good estimate. The reader will still read the estimate about the 800 gazillion dollars with skepticism, but they will be able to confirm your claim that someone made this estimate.

The point of a reference is to acknowledge where things came from. If Smith et al had the idea first, but didn't publish it, then you have to acknowledge that they came up with it first.

What you are saying here is "you'll just have to trust me on this one", and it is up to the reader/reviewer to judge if they believe it or not. Thus, you are very unlikely to see it used to support something that is important to the arguement. You are most likely to see it supporting speculation or scene setting.

The other place you might see it is "we didn't do this, because Smith et al told us they tried it and it didn't work and we didn't want to waste time and money on something that didn't work".

It is possible sometimes that a reviewer will see thing that don't make it into the final paper, but that would more likely with "(data not shown)" than "(Unpublished data)".