The past few years have seen numerous discussions raised about the pros and cons of hosting your own blog vs. riding the snake that is WordPress/Blogger/whatever. While it's safe to say that growing your own is the best practice (I think I'm still talking about blogs here...), I've never seen anyone in SEM successfully play devil's advocate on the matter. Or the "devil's avocado," as my illiterate friend used to say.

Most proponents of free-hosted blogs are recreational bloggers, not search marketers. Their list of pros revolves around ease of use and plug-n-playness; however, I'm going to assume that you're perfectly capable of installing and tweaking WP (or whatever your flavor may be) on a server, and instead are looking for reasons why you might do otherwise when working with a client.

1. The client does not have access to their website. Or, the client has to pay some agency an exorbitant amount of money every time they want to make a site change. Or, they have access to their site but claim to have incompatible frameworks (e.g., ASP.NET site vs. PHP blog). Whatever the case may be, you don't want to get in the middle of any programming b.s. or company politics if you can avoid them. Free-hosted lets you do your SEM thing without getting your hands dirty.
2. Monopolizing SERPs. We could debate whether it's better to have one site with a standard and indented listing or two sites with standard listings, but I'll assume you're capable of generating a standard and indented listing for both the client's website and free-hosted blog for certain keywords. That gives you four listings in a single SERP page, or two more than you'd likely have if the blog had been hosted as a subdirectory on the client's site. For clients with reputation management issues (or clients who just think monopolizing the SERPS is a rational goal to shoot for), the combination of 3 or 4 listings and the local results box should push any bad juju below the fold.
3. Keeps the metrics nutjobs at bay. From time to time, you'll come across a client who thinks bounce rate is the be-all end-all, or worse, someone whose salary depends upon the main website's conversion rate. Well, there are many instances where a blog's reach can extend far past the reach of the client's main website (local businesses, for instance, who are unlikely to win over a conversion from 400 miles away), and when you host a blog inside the client's site, all your traffic numbers pile into the site stats and skew the figures. Local businesses often see a drop in conversion rate when they add a blog, due to all the search-related traffic that has no geographical relevance. With a free-hosted blog, your stats remain separated, and the only traffic the main site will see is the users who have already "half-converted": they read the blog and decided they wanted to learn more about the company. Of course, you could filter your stats on a hosted setup and that would solve the problem as well, but it's usually not as clean as the aforementioned solution.
4. The client will benefit from the blog host's exclusive search traffic. The Wordpress and Blogger networks are capable of sending quite a bit of traffic to your blog via the keyword and category searches from their own user community. As far as I know, you don't get a ticket to ride that train unless you're free-hosted.

Again, I'd still recommend setting up most blogs as subdirectories of a client's main site… but it pays to know both sides of the fence if you work with a wide array of clientele.