This is the wrong question. The right question is: when does freemium work? 
The business model for Rapportive (which adds rich contact profiles to your email) is simple: build an amazingly useful product, and charge money for it. It will most likely be freemium: a base version of Rapportive will always be free, and there will be a premium version which costs money. We think that this strategy will work for us.
But why? What signals indicate whether freemium will work? I think it depends on many things: the type of product you have, the kind of company you want to build, and more.
10 questions to consider:
How do users feel about your product, as a function of time?
Products which increase in value over time are good candidates for freemium.
Neil Davidson has a good illustration of this.
Do users love your product for a short while, and then get bored of it? (casual games) Are they oblivious to it most of the time, but occasionally desperate for it? (data recovery tools) Do they always dislike it? (virus scanners) Or does it gradually get more useful over time? . For example, a tool which recovers corrupted photos from your camera is unlikely to work as freemium (cameras can corrupt files, but it happens rarely), whereas a tool such as Evernote works well as a freemium (it becomes more useful over time as you enter more data into it).
How good is your long-term retention?
With good retention, you have a better chance of converting free users to paid users.
Your product may become more useful over time, but if a large proportion of free users stop using it after a while, that's bad news if you were hoping to convert them to paying later on. In that case, you should probably ask them for money early, while you still have their attention, or improve your retention! However, if a large proportion of your free users stick around for a long time, and your product becomes more useful over time, more and more of them will eventually convert from free to paid.
Does your product require behaviour change, or can people start using it gradually?
If people can start using your product gradually, freemium might work.
For example, an image editor requires an abrupt change of behaviour: you've got to stop using the old one and start using the new one. In that case, charging all users from day one may actually help you: by having spent money, users are more likely to also spend the time learning how to use the new application (due to the sunk cost fallacy), and thus end up getting more value from the product. On the other hand, if your product doesn't have a big learning overhead and users can gently start using it, it may be better to charge for it later.
Does your product have distinct modes of use for different audiences?
If it does, you can be more creative with your freemium model.
I am fascinated by Yammer's business model. They get lots of people in an organisation to use it for free; then sometime later, when there are lots of active users, they sell the enterprise IT department on stuff like data ownership, admin controls and security tools. The end-users who sign up do not care about the admin features; the IT department does, and they are a separate audience. This is an interesting take on freemium: always free for end users, always paid-for for IT departments doing their job.
What is your market like?
If you are targeting a large unmet need, you should make your product free.
When faced with a wide open field you should be in land-grab mode, and acquire users before your competitors do. On the other hand, if you're in a well-established market, you'll need to gradually convince users to move away from their existing solutions. In this case, charging money is the best way of finding customers who care enough about the problem that they are willing to pay for a better solution.
Are you targeting a premium niche?
Free users are the opposite of premium.
Sometimes free users are more troublesome than paid users (for example, MailChimp was faced with a spam problem when they started offering a free plan). If you are going for the top end of the market, giving something away for free may hurt you more than it benefits you. The price of your product says a lot about your positioning, and people tend to assume that if something is free, it's less valuable than something expensive.
Which metric do you use to separate free from paying users?
Freemium makes sense if there's an obvious point to start charging.
Do your users need to pay when they exceed a certain number of seats, credits or widgets? Picking a metric is tricky, and is a topic worthy of a separate blog post. Number of seats is a common metric, but it only makes sense for applications where there's a downside to everyone using the same login. (For example, if your application is an analytics dashboard, it doesn't really matter if there's one shared login or each person has a separate login.) If there is no obvious metric which separates free from paying users, you should probably charge everybody.
Do you depend on word-of-mouth marketing?
More users (even if they are free) = more mouths.
If yes, note that more people using your product means more mouths to spread the word. You still need to reach the right kind of users, so the question is: can you reach people who will pay, through the word of people who will remain forever free? My guess is that if you're targeting a specific niche, word-of-mouth spread and willingness to pay are strongly correlated (which suggests that there is little benefit in having lots of free users); if you're targeting a broad audience, the two are uncorrelated, so free users can help you carry the word to people who will pay. 
What kind of company do you want to build?
You need lots of users if you want to take over the world.
If you've taken venture capital and want to take over the world, you need to grow quickly, even if it means leaving revenue on the table. If you want to grow organically and maximise profits, you're better off maximising revenue per user, and ignoring those users who would never pay you anyway.
What are your costs per user?
They had better be low if you have lots of free users.
Are your costs fixed (developers, testers) or variable (servers, support)? If your variable costs are low enough, it's fine to have a low conversion rate, because one paid-for user may pay for 1,000 free users. Support for free users is often the limiting factor. To keep your support burden low, you'll need to make your product easy to use and fix all your bugs... but that's well worth doing anyway!
So why does Ruben say that free plans don't work? Well, for his situation, I think he's absolutely right. Let's consider the above questions for Bidsketch.
Bidsketch a workflow tool for designers: it's immediately valuable and then probably stays this valuable over time. It requires a change of behaviour: stop using email and start using Bidsketch. It has two distinct modes of use (one for designers and one for clients), but it seems pitched at one audience (designers). It competes with email, a well established solution. It seems to target a premium niche of the best freelance designers; this is not a tool for everybody. Ruben is self-declared Micropreneur who wants his business to grow organically. And because he's working by himself (as far as I can tell), the support burden of free users would be significant.
So I agree: freemium is not right for Ruben's product. But it might be right for yours.
 This is a popular failure mode for online discussion, and a pet peeve of mine. 14 months or so ago, the NoSQL community was squabbling about "mine is better" – "no, mine is", so I contributed a post explaining when you should use which type of database.
 Phil Libin, CEO of Evernote, gave an excellent talk on Evernote's business metrics in May. It's packed with great insights for products which gradually become more useful over time.
"Every month, the longer you use [Evernote], the more valuable it gets. And since the long-term retention is flat, and the conversion goes up, what you see is: the longer a cohort stays, the more valuable they become." (at 14'55")
Examining the cohort of users which signed up in March 2008, Phil found that after 3 months, they were making $300/month from 11,000 people (a conversion rate of about 0.6%); 22 months later, they were making $8,000/month from the same 11,000 people (a conversion rate of about 16%)!
"Users are kind of like a nice wine, or a stinky cheese. As [a cohort of users] ages, it actually gets better. A lot of the people who wouldn't pay leave, and a lot of the people who stay end up paying. Even though there's no hard sell, and you can use Evernote forever for free, a much larger percentage winds up converting." (at 17'05")
Of course, not every application increases in value to users over time.
 Rapportive has broad appeal and is not limited to a specific niche, so it makes sense for us to have lots of free users. When Brad Feld recently said that he was trying Gmail, a chorus of our users jumped to recommend Rapportive to him. We wouldn't have got that without lots of people using and loving our product.