Add it to your site.  You want people to know the blog is part of the site, and you want people to be able to get from your site to the blog and vice versa easily.  Also, you want your site's rankings to benefit from the traffic you bring in via the blog, and vice versa.

To make it be treated as part of your site, you should set it up under a URL like mysite.com/blog, vs. blog.mysite.com.  The subdomain approach will get your blog treated like a new site.

Setting pages to 404 should be enough to remove them after Google indexes your page enough times.  Google has to be careful about this, because when many sites crash or have site maintenance, they return 404 instead of 503, so Google wouldn't want to remove pages from their index until they're sure the page is gone.

Google talks about removing pages from there index here.  The Google Webmaster Tools URL removal tool is only intended for pages that urgently need to be removed, so I wouldn't recommend that.  Google recommends:

• If the page no longer exists, make sure that the server returns a 404 (Not Found) or 410 (Gone) HTTP status code. This will tell Google that the page is gone and that it should no longer appear in search results.
• If the page still exists but you don't want it to appear in search results, use robots.txt to prevent Google from crawling it. Note that in general, even if a URL is disallowed by robots.txt we may still index the page if we find its URL on another site. However, Google won't index the page if it's blocked in robots.txt and there's an active removal request for the page.
• Alternatively, you can use a noindex meta tag. When we see this tag on a page, Google will completely drop the page from our search results, even if other pages link to it. This is a good solution if you don't have direct access to the site server. (You will need to be able to edit the HTML source of the page).

Is there a reason you are 404'ing these pages rather than redirecting them?  If these pages have new pages with similar content, you should do a 301 redirect to keep the link juice flowing and to take advantage of these pages being linked to.  If you do continue returning 404 for these pages (or even if you don't...), make sure your 404 page is a useful one, that helps users find the page they're looking for (Google help article).

Also, Ryan, I'd be interested in hearing the results of using the 410 status code.  I would imagine that status code would do the trick!  I'm surprised I haven't read about this more, or why it's not mentioned in the help file linked to above.

I like lowercase because when I type URLs by hand, I don't think to capitalize things.  If you capitalize things, you have to get the casing right to make the URL valid (unless you're setting up all sorts of fancy redirects), otherwise you a 404 and are left scratching your head.  Also, I agree with Dan that it looks better.

Hyphens vs. underscores is a classic question; Matt Cutts says to go with hyphens: http://www.youtube.com/watch?v=Q3SFVfDIS5k.  I like that better too.

301 redirect chains are bad.  If it's important to you to get the link juice to get from page A to page D, you should change it to just one redirect from A --> D.  Matt Cutts talked to Rand about it on a whiteboard Friday http://www.seomoz.org/blog/whiteboard-interview-googles-matt-cutts-on-redirects-trust-more:

Is It a Bad Idea to Chain Redirects (e.g. 301-->301-->301)?

"It is, yeah."

Matt was very clear that Google can and usually will deal with one or two redirects in a series, but three is pushing it and anything beyond that probably won't be followed.  He also reiterated that 302s should only be used for temporary redirects...but you already knew that, right?

Ryan, spot on as always.

One other thing, it sounds like some of the canonicals you're placing on pages would be better suited to 301 redirects, like correcting a URL for not having a trailing slash or not.  If you can avoid using canonicals and use 301 redirects instead, that's the preferred method for resolving duplicate content issues.  Canonicals are more for when there are parameters on the URLs, and you can't get away from serving the pages with those parameters.